Basic abstract algebra: for graduate students and advanced undergraduates

page 1 of Frontmatter Abstract Algebra: The Basic Graduate Year Robert B. Ash PREFACE This is a text for the basic gra...

Author: Robert B. Ash

105 downloads 920 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

page 1 of Frontmatter

Abstract Algebra: The Basic Graduate Year Robert B. Ash PREFACE This is a text for the basic graduate sequence in abstract algebra, oﬀered by most universities. We study fundamental algebraic structures, namely groups, rings, ﬁelds and modules, and maps between these structures. The techniques are used in many areas of mathematics, and there are applications to physics, engineering and computer science as well. In addition, I have attempted to communicate the intrinsic beauty of the subject. Ideally, the reasoning underlying each step of a proof should be completely clear, but the overall argument should be as brief as possible, allowing a sharp overview of the result. These two requirements are in opposition, and it is my job as expositor to try to resolve the conﬂict. My primary goal is to help the reader learn the subject, and there are times when informal or intuitive reasoning leads to greater understanding than a formal proof. In the text, there are three types of informal arguments: 1. The concrete or numerical example with all features of the general case. Here, the example indicates how the proof should go, and the formalization amounts to substituting Greek letters for numbers. There is no essential loss of rigor in the informal version. 2. Brief informal surveys of large areas. There are two of these, p-adic numbers and group representation theory. References are given to books accessible to the beginning graduate student. 3. Intuitive arguments that replace lengthy formal proofs which do not reveal why a result is true. In this case, explicit references to a precise formalization are given. I am not saying that the formal proof should be avoided, just that the basic graduate year, where there are many pressing matters to cope with, may not be the appropriate place, especially when the result rather than the proof technique is used in applications. I would estimate that about 90 percent of the text is written in conventional style, and I hope that the book will be used as a classroom text as well as a supplementary reference. Solutions to all problems are included in the text; in my experience, most students ﬁnd this to be a valuable feature. The writing style for the solutions is similar to that of the main text, and this allows for wider coverage as well as reinforcement of the basic ideas. Chapters 1-4 cover basic properties of groups, rings, ﬁelds and modules. The typical student will have seen some but not all of this material in an undergraduate algebra course. [It should be possible to base an undergraduate course on Chapters 1-4, traversed at a suitable pace with detailed coverage of the exercises.] In Chapter 4, the fundamental structure theorems for ﬁnitely generated modules over a principal ideal domain are developed concretely with the aid of the Smith normal form. Students will undoubtedly be comfortable with elementary row and column operations, and this will signiﬁcantly aid the learning process. In Chapter 5, the theme of groups acting on sets leads to a nice application to combinatorics as well as the fundamental Sylow theorems and some results on simple groups. Analysis of normal and subnormal series leads to the Jordan-H¨ older theorem and to solvable and nilpotent groups. The ﬁnal section, on deﬁning a group by generators and relations, concentrates on practical cases where the structure of a group can be deduced from its presentation. Simplicity of the alternating groups and semidirect products are covered in the exercises. Chapter 6 goes quickly to the fundamental theorem of Galois theory; this is possible because the necessary background has been covered in Chapter 3. After some examples of

page 2 of Frontmatter

direct calculation of a Galois group, we proceed to ﬁnite ﬁelds, which are of great importance in applications, and cyclotomic ﬁelds, which are fundamental in algebraic number theory. The Galois group of a cubic is treated in detail, and the quartic is covered in an appendix. Sections on cyclic and Kummer extensions are followed by Galois’ fundamental theorem on solvability by radicals. The last section of the chapter deals with transcendental extensions and transcendence bases. In the remaining chapters, we begin to apply the results and methods of abstract algebra to related areas. The title of each chapter begins with “Introducing. . . ”, and the areas to be introduced are algebraic number theory, algebraic geometry, noncommutative algebra and homological algebra (including categories and functors). Algebraic number theory and algebraic geometry are the two major areas that use the tools of commutative algebra (the theory of commutative rings). In Chapter 7, after an example showing how algebra can be applied in number theory, we assemble some algebraic equipment: integral extensions, norms, traces, discriminants, Noetherian and Artinian modules and rings. We then prove the fundamental theorem on unique factorization of ideals in a Dedekind domain. The chapter concludes with an informal introduction to p-adic numbers and some ideas from valuation theory. Chapter 8 begins geometrically with varieties in aﬃne space. This provides motivation for Hilbert’s fundamental theorems, the basis theorem and the nullstellensatz. Several equivalent versions of the nullstellensatz are given, as well as some corollaries with geometric signiﬁcance. Further geometric considerations lead to the useful algebraic techniques of localization and primary decomposition. The remainder of the chapter is concerned with the tensor product and its basic properties. Chapter 9 begins the study of noncommutative rings and their modules. The basic theory of simple and semisimple rings and modules, along with Schur’s lemma and Jacobson’s theorem, combine to yield Wedderburn’s theorem on the structure of semisimple rings. We indicate the precise connection between the two popular deﬁnitions of simple ring in the literature. After an informal introduction to group representations, Maschke’s theorem on semisimplicity of modules over the group algebra is proved. The introduction of the Jacobson radical gives more insight into the structure of rings and modules. The chapter ends with the Hopkins-Levitzki theorem that an Artinian ring is Noetherian, and the useful lemma of Nakayama. In Chapter 10, we introduce some of the tools of homological algebra. Waiting until the last chapter for this is a deliberate decision. Students need as much exposure as possible to speciﬁc algebraic systems before they can appreciate the broad viewpoint of category theory. Even experienced students may have diﬃculty absorbing the abstract deﬁnitions of kernel, cokernel, product, coproduct, direct and inverse limit. To aid the reader, functors are introduced via the familiar examples of hom and tensor. No attempt is made to work with general abelian categories. Instead, we stay within the category of modules and study projective, injective and ﬂat modules. In a supplement, we go much farther into homological algebra than is usual in the basic algebra sequence. We do this to help students cope with the massive formal machinery that makes it so diﬃcult to gain a working knowledge of this area. We concentrate on the results that are most useful in applications: the long exact homology sequence and the properties of the derived functors Tor and Ext. There is a complete proof of the snake lemma, a rarity in the literature. In this case, going through a long formal proof is entirely appropriate, because doing so will help improve algebraic skills. The point is not to avoid diﬃculties, but to make most eﬃcient use of the ﬁnite amount of time available. Robert B. Ash October 2000 Further Remarks Many mathematicians believe that formalism aids understanding, but I believe that

page 3 of Frontmatter

when one is learning a subject, formalism often prevents understanding. The most important skill is the ability to think intuitively. This is true even in a highly abstract ﬁeld such as homological algebra. My writing style reﬂects this view. Classroom lectures are inherently ineﬃcient. If the pace is slow enough to allow comprehension as the lecture is delivered, then very little can be covered. If the pace is fast enough to allow decent coverage, there will unavoidably be large gaps. Thus the student must depend on the textbook, and the current trend in algebra is to produce massive encyclopedias, which are likely to be quite discouraging to the beginning graduate student. Instead, I have attempted to write a text of manageable size, which can be read by sudents, including those working independently. Another goal is to help the student reach an advanced level as quickly and eﬃciently as possible. When I omit a lengthy formal argument, it is because I judge that the increase in algebraic skills is insuﬃcient to justify the time and eﬀort involved in going through the formal proof. In all cases, I give explicit references where the details can be found. One can argue that learning to write formal proofs is an essential part of the student’s mathematical training. I agree, but the ability to think intuitively is fundamental and must come ﬁrst. I would add that the way things are today, there is absolutely no danger that the student will be insuﬃciently exposed to formalism and abstraction. In fact there is quite a bit of it in this book, although not 100 percent. I oﬀer this text in the hope that it will make the student’s trip through algebra more enjoyable. I have done my best to avoid gaps in the reasoning. I never use the phrase “it is easy to see” under any circumstances. I welcome comments and suggestions for improvement. c 2000, by Robert B. Ash Copyright Paper or electronic copies for noncommercial use may be made freely without explicit permission of the author. All other rights are reserved.

page 4 of Frontmatter

ABSTRACT ALGEBA: THE BASIC GRADUATE YEAR TABLE OF CONTENTS CHAPTER 0 PREREQUISITES 0.1 Elementary Number Theory 0.2 Set Theory 0.3 Linear Algebra CHAPTER 1 GROUP FUNDAMENTALS 1.1 Groups and Subgroups 1.2 Permutation Groups 1.3 Cosets, Normal Subgroups and Homomorphisms 1.4 The Isomorphism Theorems 1.5 Direct Products CHAPTER 2 RING FUNDAMENTALS 2.1 Basic Deﬁnitions and Properties 2.2 Ideals, Homomorphisms and Quotient Rings 2.3 The Isomorphism Theorems For Rings 2.4 Maximal and Prime Ideals 2.5 Polynomial Rings 2.6 Unique Factorization 2.7 Principal Ideal Domains and Euclidean Domains 2.8 Rings of Fractions 2.9 Irreducible Polynomials CHAPTER 3 FIELD FUNDAMENTALS 3.1 Field Extensions 3.2 Splitting Fields 3.3 Algebraic Closures 3.4 Separability 3.5 Normal Extensions CHAPTER 4 MODULE FUNDAMENTALS 4.1 Modules and Algebras 4.2 The Isomorphism Theorems For Modules 4.3 Direct Sums and Free Modules 4.4 Homomorphisms and Matrices 4.5 Smith Normal Form 4.6 Fundamental Structure Theorems 4.7 Exact Sequences and Diagram Chasing CHAPTER 5 SOME BASIC TECHNIQUES OF GROUP THEORY 5.1 Groups Acting on Sets 5.2 The Orbit-Stabilizer Theorem 5.3 Applications to Combinatorics 5.4 The Sylow Theorems 5.5 Applications of the Sylow Theorems 5.6 Composition Series 5.7 Solvable and Nilpotent Groups

page 5 of Frontmatter

5.8 Generators and Relations CHAPTER 6 GALOIS THEORY 6.1 Fixed Fields and Galois Groups 6.2 The Fundamental Theorem 6.3 Computing a Galois Group Directly 6.4 Finite Fields 6.5 Cyclotomic Fields 6.6 The Galois Group of a Cubic 6.7 Cyclic and Kummer Extensions 6.8 Solvability by Radicals 6.9 Transcendental Extensions Appendix to Chapter 6 CHAPTER 7 INTRODUCING ALGEBRAIC NUMBER THEORY 7.1 Integral Extensions 7.2 Quadratic Extensions of the Rationals 7.3 Norms and Traces 7.4 The Discriminant 7.5 Noetherian and Artinian Modules and Rings 7.6 Fractional Ideals 7.7 Unique Factorization of Ideals in a Dedekind Domain 7.8 Some Arithmetic in Dedekind Domains 7.9 p-adic Numbers CHAPTER 8 INTRODUCING ALGEBRAIC GEOMETRY 8.1 Varieties 8.2 The Hilbert Basis Theorem 8.3 The Nullstellensatz: Preliminaries 8.4 The Nullstellensatz: Equivalent Versions and Proof 8.5 Localization 8.6 Primary Decomposition 8.7 Tensor Product of Modules Over a Commutative Ring 8.8 General Tensor Products CHAPTER 9 INTRODUCING NONCOMMUTATIVE ALGEBRA 9.1 Semisimple Modules 9.2 Two Key Theorems 9.3 Simple and Semisimple Rings 9.4 Further Properties of Simple Rings, Matrix Rings, and Endomorphisms 9.5 The Structure of Semisimple Rings 9.6 Maschke’s Theorem 9.7 The Jacobson Radical 9.8 Theorems of Hopkins-Levitzki and Nakayama CHAPTER 10 INTRODUCING HOMOLOGICAL ALGEBRA 10.1 Categories 10.2 Products and Coproducts 10.3 Functors 10.4 Exact Functors 10.5 Projective Modules 10.6 Injective Modules

page 6 of Frontmatter

10.7 Embedding into an Injective Module 10.8 Flat Modules 10.9 Direct and Inverse Limits Appendix to Chapter 10 SUPPLEMENT S1 Chain Complexes S2 The Snake Lemma S3 The Long Exact Homology Sequence S4 Projective and Injective Resolutions S5 Derived Functors S6 Some Properties of Ext and Tor S7 Base Change in the Tensor Product SOLUTIONS TO PROBLEMS

page 1 of Chapter 0

Chapter 0 PREREQUISITES All topics listed in this chapter are covered in A Primer of Abstract Mathematics by Robert B. Ash, MAA 1998. 0.1 Elementary Number Theory The greatest common divisor of two integers can be found by the Euclidean algorithm, which is reviewed in the exercises in Section 2.5. Among the important consequences of the algorithm are the following three results. 0.1.1 If d is the greatest common divisor of a and b, then there are integers s and t such that sa + tb = d. In particular, if a and b are relatively prime, there are integers s and t such that sa + tb = 1. 0.1.2 If a prime p divides a product a1 · · · an of integers, then p divides at least one ai 0.1.3 Unique Factorization Theorem If a is an integer, not 0 or ±1, then (1) a can be written as a product p1 · · · pn of primes. (2) If a = p1 · · · pn = q1 · · · qm , where the pi and qj are prime, then n = m and, after renumbering, pi = ±qi for all i. [We allow negative primes, so that, for example, -17 is prime. This is consistent with the general deﬁnition of prime element in an integral domain; see Section 2.6.] 0.1.4 The Integers Modulo m If a and b are integers and m is a positive integer ≥ 2, we write a ≡ b mod m, and say that a is congruent to b modulo m, if a − b is divisible by m. Congruence modulo m is an equivalence relation, and the resulting equivalence classes are called residue classes mod m. Residue classes can be added, subtracted and multiplied consistently by choosing a representative from each class, performing the appropriate operation, and calculating the residue class of the result. The collection Zm of residue classes mod m forms a commutative ring under addition and multiplication. Zm is a ﬁeld if and only if m is prime. (The general deﬁnitions of ring, integral domain and ﬁeld are given in Section 2.1.) 0.1.5 (1) The integer a is relatively prime to m if and only if a is a unit mod m, that is, a has a multiplicative inverse mod m. (2) If c divides ab and a and c are relatively prime, then c divides b. (3) If a and b are relatively prime to m, then ab is relatively prime to m. (4) If ax ≡ ay mod m and a is relatively prime to m, then x ≡ y mod m. (5) If d = gcd(a, b), the greatest common divisor of a and b, then a/d and b/d are relatively prime. (6) If ax ≡ ay mod m and d = gcd(a, m), then x ≡ y mod m/d. (7) If ai divides b for i = 1, . . . , r, and ai and aj are relatively prime whenever i = j, then the product a1 · · · ar divides b. (8) The product of two integers is their greatest common divisor times their least common multiple. 0.1.6 Chinese Remainder Theorem If m1 , . . . , mr are relatively prime in pairs, then the system of simultaneous equations x ≡ bj mod mj , j = 1, . . . , r, has a solution for arbitrary integers bj . The set of solutions forms a single residue class mod m=m1 · · · mr , so that there is a unique solution mod m. This result can be derived from the abstract form of the Chinese remainder theorem; see Section 2.3. 0.1.7 Euler’s Theorem The Euler phi function is deﬁned by ϕ(n) = the number of integers in {1, . . . , n} that are relatively prime to n. For an explicit formula for ϕ(n), see

page 2 of Chapter 0

Section 1.1, Problem 13. Euler’s theorem states that if n ≥ 2 and a is relatively prime to n, then aϕ(n) ≡ 1 mod n. 0.1.8 Fermat’s Little Theorem If a is any integer and p is a prime not dividing a, then ap−1 ≡ 1 mod p. Thus for any integer a and prime p, whether or not p divides a, we have ap ≡ a mod p. For proofs of (0.1.7) and (0.1.8), see (1.3.4). 0.2 Set Theory 0.2.1 A partial ordering on a set S is a relation on S that is reﬂexive (x ≤ x for all x ∈ S), antisymmetric (x ≤ y and y ≤ x implies x = y), and transitive (x ≤ y and y ≤ z implies x ≤ z). If for all x, y ∈ S, either x ≤ y or y ≤ x, the ordering is total. 0.2.2 A well-ordering on S is a partial ordering such that every nonempty subset A of S has a smallest element a. (Thus a ≤ b for every b ∈ A). 0.2.3 Well-Ordering Principle Every set can be well-ordered. 0.2.4 Maximum Principle If T is any chain (totally ordered subset) of a partially ordered set S, then T is contained in a maximal chain M . (Maximal means that M is not properly contained in a larger chain.) 0.2.5 Zorn’s Lemma If S is a nonempty partially ordered set such that every chain of S has an upper bound in S, then S has a maximal element. (The element x is an upper bound of the set A if a ≤ x for every a ∈ A. Note that x need not belong to A, but in the statement of Zorn’s lemma, we require that if A is a chain of S, then A has an upper bound that actually belongs to S.) 0.2.6 Axiom of Choice Given any family of nonempty sets Si , i ∈ I, we can choose an element of each Si . Formally, there is a function f whose domain is I such that f (i) ∈ Si for all i ∈ I. The well-ordering principle, the maximum principle, Zorn’s lemma, and the axiom of choice are equivalent in the sense that if any one of these statements is added to the basic axioms of set theory, all the others can be proved. The statements themselves cannot be proved from the basic axioms. Constructivist mathematics rejects the axiom of choice and its equivalents. In this philosophy, an assertion that we can choose an element from each Si must be accompanied by an explicit algorithm. The idea is appealing, but its acceptance results in large areas of interesting and useful mathematics being tossed onto the scrap heap. So at present, the mathematical mainstream embraces the axiom of choice, Zorn’s lemma et al. 0.2.7 Proof by Transﬁnite Induction To prove that statement Pi holds for all i in the well-ordered set I, we do the following: 1. Prove the basis step P0 , where 0 is the smallest element of I. 2. If i > 0 and we assume that Pj holds for all j < i (the transﬁnite induction hypothesis), prove Pi . It follows that Pi is true for all i. 0.2.8 We say that the size of the set A is less than or equal to the size of B (notation A ≤s B) if there is an injective map from A to B. We say that A and B have the same size (A =s B) if there is a bijection between A and B. 0.2.9 Schr¨ oder-Bernstein Theorem If A ≤s B and B ≤s A, then A =s B. (This can be proved without the axiom of choice.) 0.2.10 Using (0.2.9), one can show that if sets of the same size are called equivalent, then ≤s on equivalence classes is a partial ordering. It follows with the aid of Zorn’s lemma that

page 3 of Chapter 0

the ordering is total. The equivalence class of a set A, written |A|, is called the cardinal number or cardinality of A. In practice, we usually identify |A| with any convenient member of the equivalence class, such as A itself. 0.2.11 For any set A, we can always produce a set of greater cardinality, namely the power set 2A , that is, the collection of all subsets of A. 0.2.12 Deﬁne addition and multiplication of cardinal numbers by |A| + |B| = |A ∪ B| and |A||B| = |A × B|. In deﬁning addition, we assume that A and B are disjoint. (They can always be disjointized by replacing a ∈ A by (a, 0) and b ∈ B by (b, 1).) 0.2.13 If ℵ0 is the cardinal number of a countably inﬁnite set, then ℵ0 + ℵ0 = ℵ0 ℵ0 = ℵ0 . More generally, (a) If α and β are cardinals, with α ≤ β and β inﬁnite, then α + β = β. (b) If α = 0 (i.e., α is nonempty), α ≤ β and β is inﬁnite, then αβ = β. 0.2.14 If A is an inﬁnite set, then A and the set of all ﬁnite subsets of A have the same cardinality. 0.3 Linear Algebra It is not feasible to list all results presented in an undergraduate course in linear algebra. Instead, here is a list of topics that are covered in a typical course. 1. Sums, products, transposes, inverses of matrices; symmetric matrices. 2. Elementary row and column operations; reduction to echelon form. 3. Determinants: evaluation by Laplace expansion and Cramer’s rule. 4. Vector spaces over a ﬁeld; subspaces, linear independence and bases. 5. Rank of a matrix; homogeneous and nonhomogeneous linear equations. 6. Null space and range of a matrix; the dimension theorem. 7. Linear transformations and their representation by matrices. 8. Coordinates and matrices under change of basis. 9. Inner product spaces and the projection theorem. 10. Eigenvalues and eigenvectors; diagonalization of matrices with distinct eigenvalues, symmetric and Hermitian matrices. 11. Quadratic forms. 12. 13. 14. 15. 16.

A more advanced course might cover the following topics: Generalized eigenvectors and the Jordan canonical form. The minimal and characteristic polynomials of a matrix; Cayley-Hamilton theorem. The adjoint of a linear operator. Projection operators. Normal operators and the spectral theorem.

page 1 of Chapter 1

CHAPTER 1 GROUP FUNDAMENTALS 1.1 Groups and Subgroups 1.1.1 Deﬁnition A group is a nonempty set G on which there is deﬁned a binary operation (a, b) → ab satisfying the following properties: Closure: If a and b belong to G, then ab is also in G; Associativity: a(bc) = (ab)c for all a, b, c ∈ G; Identity: There is an element 1 in G such that a1 = 1a = a for all a in G; Inverse: If a is in G there is an element a−1 in G such that aa−1 = a−1 a = 1. A group G is abelian if the binary operation is commutative, i.e., ab = ba for all a, b in G. In this case the binary operation is often written additively ((a, b) → a + b)), with the identity written as 0 rather than 1. There are some very familiar examples of abelian groups under addition, namely the integers Z, the rationals Q, the real numbers R, the complex numbers C, and the integers Zm modulo m. Nonabelian groups will begin to appear in the next section. The associative law generalizes to products of any ﬁnite number of elements, for example, (ab)(cde) = a(bcd)e. A formal proof can be given by induction: if two people A and B form a1 · · · an in diﬀerent ways, the last multiplication performed by A might look like (a1 · · · ai )(ai+1 · · · an ), and the last multiplication by B might be (a1 · · · aj )(aj+1 · · · an ). But if (without loss of generality) i < j, then (induction hypothesis) (a1 · · · aj ) = (a1 · · · ai )(ai+1 · · · aj ) and (ai+1 · · · an ) = (ai+1 · · · aj )(aj+1 · · · an ). By the n = 3 case, i.e., the associative law as stated in the deﬁnition of a group, the products computed by A and B are the same. The identity is unique (1 = 1 1 = 1), as is the inverse of any given element (if b and b are inverses of a then b = 1b = (b a)b = b (ab) = b 1 = b ). Exactly the same argument shows that if b is a right inverse, and b a left inverse, of a, then b = b . 1.1.2 Deﬁnitions and Comments A subgroup H of a group G is a nonempty subset of G that forms a group under the binary operation of G. Equivalently, H is a nonempty subset of G such that if a and b belong to H, so does ab−1 . (Note that 1 = aa−1 ∈ H; also ab = a((b−1 )−1 ) ∈ H.) If A is any subset of a group G, the subgroup generated by A is the smallest subgroup containing A, often denoted by < A >. Formally, < A > is the intersection of all subgroups containing A. More explicitly, < A > consists of all ﬁnite products a1 · · · an , n = 1, 2, ..., where for each i, either ai or a−1 belongs to A. (All such products belong to any subgroup i containing A, and the collection of all such products forms a subgroup. In checking that the inverse of an element of < A > also belongs to < A >, we use the fact that −1 (a1 · · · an )−1 = a−1 n · · · a1 −1 which is veriﬁed directly:(a1 · · · an )(a−1 n · · · a1 ) = 1.)

1.1.3 Deﬁnitions and Comments The groups G1 and G2 are said to be isomorphic if there is a bijection f : G1 → G2 that preserves the group operation, in other words, f (ab) = f (a)f (b). Isomorphic groups are essentially the same; they diﬀer only notationally. Here is a simple example. A group G is cyclic if G is generated by a single element: G =< a >. A ﬁnite cyclic group generated by a is necessarily abelian, and can be written as {1, a, a2 , ..., an−1 } where an = 1, or in additive notation, {0, a, 2a, ..., (n − 1)a}, with na = 0. Thus a ﬁnite cyclic group with n elements is isomorphic to the additive group Zn of integers

page 2 of Chapter 1

modulo n. Similarly, if G is an inﬁnite cyclic group generated by a, then G must be abelian and can be written as {1, a, a2 , ...}, or in additive notation, {0, a, 2a, ...}. In this case, G is isomorphic to the additive group Z of all integers. The order of an element in a group G (notation |a|) is the least positive integer n such that an = 1; if no such integer exists, the order of a is inﬁnite. Thus if |a| = n, then the cyclic subgroup < a > generated by a has exactly n elements, and ak = 1 iﬀ k is a multiple of n. (Concrete examples are more illuminating than formal proofs here. Start with 0 in the integers modulo 4, and continually add 1; the result is 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3....) The order of the group G, denoted by |G|, is simply the number of elements in G 1.1.4 Proposition If G is a ﬁnite cyclic group of order n, then G has exactly one (necessarily cyclic) subgroup of order n/d for each positive divisor d of n, and G has no other subgroups. If G is an inﬁnite cyclic group, the (necessarily cyclic) subgroups of G are of the form {1, b, b2 , ...}, where b is an arbitrary element of G, or in additive notation, {0, b, 2b, ...}. Proof. Again, an informal argument is helpful. Suppose that H is a subgroup of Z20 (the integers with addition modulo 20). If the smallest positive integer in H is 6 (a non-divisor of 20) then H contains 6, 12, 18, 4 (oops, a contradiction, 6 is supposed to be the smallest positive integer). On the other hand, if the smallest positive integer in H is 4, then H = {4,8,12,16,0}. Similarly, if the smallest positive integer in a subgroup H of the additive group of integers Z is 5, then H = {0, 5, 10, 15, 20, ...}.♣ If G = {1, a, ..., an−1 } is a cyclic group of order n, when will an element ar also have order n? To discover the answer, let’s work in Z12 . Does 8 have order 12? We compute 8, 16, 24(= 0), so the order of 8 is 3. But if we try 7, we get 7, 14, 21, 28, 35, 42, 49, 56, 63, 70, 77, 84 = 7 × 12, so 7 does have order 12. The point is that the least common multiple of 7 and 12 is simply the product, while the lcm of 8 and 12 is smaller than the product. Equivalently, the greatest common divisor of 7 and 12 is 1, while the gcd of 8 and 12 is 4 > 1. We have the following result. 1.1.5 Proposition If G is a cyclic group of order n generated by a, the following conditions are equivalent: (a) |ar | = n. (b) r and n are relatively prime. (c) r is a unit mod n, in other words, r has an inverse mod n (an integer s such that rs ≡ 1 mod n). Furthermore, the set Un of units mod n forms a group under multiplication. The order of this group is ϕ(n) = the number of positive integers less than or equal to n that are relatively prime to n; ϕ is the familiar Euler ϕ function. Proof. The equivalence of (a) and (b) follows from the discussion before the statement of the proposition, and the equivalence of (b) and (c) is handled by a similar argument. For example, since there are 12 distinct multiples of 7 mod 12, one of them must be 1; speciﬁcally, 7 × 7 ≡ 1 mod 12. But since 8 × 3 is 0 mod 12, no multiple of 8 can be 1 mod 12. (If 8x ≡ 1, multiply by 3 to reach a contradiction.). Finally, Un is a group under multiplication because the product of two integers relatively prime to n is also relatively prime to n.♣ Problems For Section 1.1 1. A semigroup is a nonempty set with a binary operation satisfying closure and associativity (we drop the identity and inverse properties from the deﬁnition of a group).A monoid is a semigroup with identity (so that only the inverse property is dropped). Give an example of a monoid that is not a group, and an example of a semigroup that is not a monoid. 2. In Z6 , the group of integers modulo 6, ﬁnd the order of each element. 3. List all subgroups of Z6 .

page 3 of Chapter 1

4. Let S be the set of all n by n matrices with real entries. Does S form a group under matrix addition? 5. Let S ∗ be the set of all nonzero n by n matrices with real entries. Does S ∗ form a group under matrix multiplication? 6. If H is a subgroup of the integers Z and H = {0}, what does H look like? 7. Give an example of an inﬁnite group that has a nontrivial ﬁnite subgroup (trivial means consisting of the identity alone). 8. Let a and b belong to the group G. If ab = ba and |a| = m, |b| = n, where m and n are relatively prime, show that |ab| = mn and that < a > ∩ = {1}. 9. If G is a ﬁnite abelian group, show that G has an element g such that |g| is the least common multiple of {|a| : a ∈ G}. 10. Show that a group G cannot be the union of two proper subgroups, in other words, if G = H ∪ K where H and K are subgroups of G, then H = G or K = G. Equivalently, if H and K are subgroups of a group G, then H ∪ K cannot be a subgroup unless H ⊆ K or K ⊆ H. 11. In an arbitrary group, let a have ﬁnite order n, and let k be a positive integer. If (n, k) is the greatest common divisor of n and k, and [n, k] the least common multiple, show that the order of ak is n/(n, k) = [n, k]/k. 12. Suppose that the prime factorization of the positive integer n is n = pe11 pe22 · · · perr and let Ai be the set of all positive integers m ∈ {1, 2, ..., n} such that pi divides m. Show that if |S| is the number of elements in the set S, then |Ai | =

n n n , |Ai ∩ Aj | = for i = j, |Ai ∩ Aj ∩ Ak | = for i, j, k distinct, pi p i pj pi pj pk

and so on. 13. Continuing Problem 12, show that the number of positive integers less than or equal to n that are relatively prime to n is ϕ(n) = n(1 −

1 1 1 )(1 − ) · · · (1 − ) p1 p2 pr

. 14. Give an example of a ﬁnite group G (of order at least 3) such that the only subgroups of G are {1} and G itself. 15. Does an inﬁnite group with this property exist? 1.2 Permutation Groups 1.2.1 Deﬁnition A permutation of a set S is a bijection on S, that is, a function π : S → S that is one-to-one and onto. (If S is ﬁnite, then π is one-to-one if and only if it is onto.) If S is not too large, it is feasible to describe a permutation by listing the elements x ∈ S and the corresponding values π(x). For example, if S = {1, 2, 3, 4, 5}, then π=

1 3

2 5

3 4

4 1

5 2

is the permutation such that π(1) = 3, π(2) = 5, π(3) = 4, π(4) = 1, π(5) = 2. If we start with any element x ∈ S and apply π repeatedly to obtain π(x), π(π(x)), π(π(π(x))), and so on, eventually we must return to x, and there are no repetitions along

page 4 of Chapter 1

the way because π is one-to-one. For the above example, we obtain 1 → 3 → 4 → 1, 5 → 2. We express this result by writing

2→

π = (1, 3, 4)(2, 5) where the cycle (1,3,4) is the permutation of S that maps 1 to 3, 3 to 4 and 4 to 1, leaving the remaining elements 2 and 5 ﬁxed. Similarly, (2,5) maps 2 to 5, 5 to 2, 1 to 1, 3 to 3 and 4 to 4. The product of (1,3,4) and (2,5) is interpreted as a composition, with the right factor (2,5) applied ﬁrst, as with composition of functions. In this case, the cycles are disjoint, so it makes no diﬀerence which mapping is applied ﬁrst. The above analysis illustrates the fact that any permutation can be expressed as a product of disjoint cycles, and the cycle decomposition is unique. 1.2.2 Deﬁnitions and Comments A permutation π is said to be even if its cycle decomposition contains an even number of even cycles (that is, cycles of even length); otherwise π is odd. A cycle can be decomposed further into a product of (not necessarily disjoint) twoelement cycles, called transpositions. For example, (1, 2, 3, 4, 5) = (1, 5)(1, 4)(1, 3)(1, 2) where the order of application of the mappings is from right to left. Multiplication by a transposition changes the parity of a permutation (from even to odd, or vice versa). For example, (2, 4)(1, 2, 3, 4, 5) = (2, 3)(1, 4, 5) (2, 6)(1, 2, 3, 4, 5) = (1, 6, 2, 3, 4, 5); (1,2,3,4,5) has no cycles of even length so is even; (2,3)(1,4,5) and (1,6,2,3,4,5) each have one cycle of even length so are odd. Since a cycle of even length can be expressed as the product of an odd number of transpositions, we can build an even permutation using an even number of transpositions, and an odd permutation requires an odd number of transpositions. A decomposition into transpositions is not unique, for example, (1,2,3,4,5) = (1,4)(1,5)(1,4)(1,3)(1,2)(3,5), but as mentioned above, the cycle decomposition is unique. Since multiplication by a transposition changes the parity, it follows that if a permutation is expressed in two diﬀerent ways as a product of transpositions, the number of transpositions will agree in parity (both even or both odd). Consequently, the product of two even permutations is even; the product of two odd permutations is even; and the product of an even and an odd permutation is odd. To summarize very compactly, deﬁne the sign of the permutation π as sgn(π) =

+1 if π is even −1 if π is odd

Then for arbitrary permutations π1 and π2 we have sgn(π1 π2 ) = sgn(π1 )sgn(π2 ) 1.2.3 Deﬁnitions and Comments There are several permutation groups that are of major interest. The set Sn of all permutations of {1, 2, . . . , n} is called the symmetric group on n letters, and its subgroup An of all even permutations of {1, 2, . . . , n} is called the alternating group on n letters. (The group operation is composition of functions.) Since there are as many even permutations as odd ones (any transposition, when applied to the

page 5 of Chapter 1

members of Sn , produces a one-to-one correspondence between even and odd permutations), it follows that An is half the size of Sn . Denoting the size of the set S by |S|, we have |Sn | = n!,

|An | = 12 n!

We now deﬁne and discuss informally D2n , the dihedral group of order 2n. Consider a regular polygon with center O and vertices V1 , V2 , . . . , Vn , arranged so that as we move counterclockwise around the ﬁgure, we encounter V1 , V2 , . . . in turn. To eliminate some of the abstraction, let’s work with a regular pentagon with vertices A, B, C, D, E, as shown in Figure 1.2.1.

C B

D O E

A

Figure 1.2.1

The group D10 consists of the symmetries of the pentagon, i.e., those permutations that can be realized via a rigid motion (a combination of rotations and reﬂections). All symmetries can be generated by two basic operations: R = counterclockwise rotation by

360 n

=

360 5

= 72 degrees,

F (“ﬂip”) = reﬂection about the line joining the center O to the ﬁrst vertex (A in this case). The group D2n contains 2n elements, namely, I (the identity), R, R2 , . . . , Rn−1 , F , RF , R2 F , . . . , Rn−1 F (RF means F followed by R). For example, in the case of the pentagon, F = (B, E)(C, D) and R = (A, B, C, D, E), so RF = (A, B)(C, E), which is the reﬂection about the line joining O to D; note that RF can also be expressed as F R−1 . In visualizing the eﬀect of a permutation such as F , interpret F ’s taking B to E as vertex B moving to where vertex E was previously. D2n will contain exactly n rotations I, R, . . . , Rn−1 and n reﬂections F, RF, . . . , Rn−1 F . If n is odd, each reﬂection is determined by a line joining the center to a vertex (and passing through the midpoint of the opposite side). If n is even, half the reﬂections are determined by a line passing through two vertices (as well as the center), and the other half by a line passing through the midpoints of two opposite sides (as well as the center). 1.2.4 An Abstract Characterization of the Dihedral Group Consider the free group with generators R and F , in other words all ﬁnite sequences whose components are R, R−1 , F and F −1 . The group operation is concatenation, subject to the constraint that if a symbol and its inverse occur consecutively, they may be cancelled. For example, RF F F F −1 RF R−1 RF F is identiﬁed with RF F RF F F , also written as RF 2 RF 3 . If we add further restrictions (so the group is no longer “free”), we can obtain D2n . Speciﬁcally, D2n is the group deﬁned by generators R and F , subject to the relations Rn = I,

F 2 = I, and RF = F R−1 .

The relations guarantee that there are only 2n distinct group elements I, R, . . . , Rn−1 and F, RF, . . . , Rn−1 F . For example, with n = 5 we have F 2 R2 F = F F RRF = F F RF R−1 = F F F R−1 R−1 = F R−2 = F R3 ;

page 6 of Chapter 1

also, R cannot be the same as R2 F , since this would imply that I = RF , or F = R−1 = R4 , and there is no way to get this using the relations. Since the product of two group elements is completely determined by the deﬁning relations, it follows that there cannot be more than one group with the given generators and relations. (This statement is true ”up to isomorphism”; it is always possible to create lots of isomorphic copies of any given group.) The symmetries of the regular n-gon provide a concrete realization. Later we will look at more systematic methods of analyzing groups deﬁned by generators and relations. Problems For Section 1.2 1. Find the cycle decomposition of the permutation 1 2 3 4 5 6 4 6 3 1 2 5 and determine whether the permutation is even or odd. 2. Consider the dihedral group D8 as a group of permutations of the square. Assume that as we move counterclockwise around the square, we encounter the vertices A, B, C, D in turn. List all the elements of D8 . 3. In S5 , how many 5-cycles are there, i.e., how many permutations are there with the same cycle structure as (1,2,3,4,5)? 4. In S5 , how many permutations are products of two disjoint transpositions, such as (1,2)(3,4)? 5. Show that if n ≥ 3, then Sn is not abelian. 6. Show that the products of two disjoint transpositions in S4 , together with the identity, form an abelian subgroup V of S4 . Describe the multiplication table of V (known as the four group). 7. Show that the cycle structure of the inverse of a permutation π coincides with that of π. In particular, the inverse of an even permutation is even (and the inverse of an odd permutation is odd), so that An is actually a group. 8. Find the number of 3-cycles, i.e., permutations consisting of exactly one cycle of length 3, in S4 . 9. Suppose that H is a subgroup of A4 with the property that for every permutation π in A4 , π 2 belongs to H. Show that H contains all 3-cycles in A4 . (Since 3-cycles are even, H in fact contains all 3-cycles in S4 .) 10. Consider the permutation 1 2 3 4 5 π= 2 4 5 1 3 Count the number of inversions of π, that is, the number of pairs of integers that are out of their natural order in the second row of π. For example, 2 and 5 are in natural order, but 4 and 3 are not. Compare your result with the parity of π. 11. Show that the parity of any permutation π is the same as the parity of the number of inversions of π. 1.3 Cosets, Normal Subgroups, and Homomorphisms 1.3.1 Deﬁnitions and Comments Let H be a subgroup of the group G. If g ∈ G, the right coset of H generated by g is Hg = {hg : h ∈ H}; similarly, the left coset of H generated by g is gH = {gh : h ∈ H}.

page 7 of Chapter 1

It follows from the deﬁnitions (Problem 1) that if a, b ∈ G, then Ha = Hb if and only if ab−1 ∈ H, and aH = bH if and only if a−1 b ∈ H. Thus if we deﬁne a and b to be equivalent iﬀ ab−1 ∈ H, we have an equivalence relation (Problem 2), and the equivalence class of a is (Problem 3) {b : ab−1 ∈ H} = Ha. Therefore the right cosets partition G(similarly for the left cosets). Since h → ha, h ∈ H, is a one-to-one correspondence, each coset has |H| elements. There are as many right cosets as left cosets, since the map aH → Ha−1 is a one-to-one correspondence (Problem 4). If [G : H], the index of H in G, denotes the number of right (or left) cosets, we have the following basic result. 1.3.2 Lagrange’s Theorem If H is a subgroup of G, then |G| = |H|[G : H]. In particular, if G is ﬁnite then |H| divides |G|, and |G| = [G : H]. |H| Proof. There are [G : H] cosets, each with |H| members. ♣ 1.3.3 Corollary

Let G be a ﬁnite group.

(i) If a ∈ G then |a| divides |G|; in particular, a|G| = 1. Thus |G| is a multiple of the order of each of its elements, so if we deﬁne the exponent of G to be the least common multiple of {|a| : a ∈ G}, then |G| is a multiple of the exponent. (ii) If G has prime order, then G is cyclic. Proof. If the element a ∈ G has order n, then H = {1, a, a2 , ..., an−1 }is a cyclic subgroup of G with |H| = n. By Lagrange’s theorem, n divides |G|, proving (i). If |G| is prime then we may take a = 1, and consequently n = |G|. Thus H is a subgroup with as many elements as G, so in fact H and G coincide, proving (ii). ♣ Here is another corollary. 1.3.4 Euler’s Theorem If a and n are relatively prime positive integers, with n ≥ 2, then aϕ(n) ≡ 1 mod n. A special case is Fermat’s Little Theorem: if p is a prime and a is a positive integer not divisible by p, then ap−1 ≡ 1 mod p. Proof. The group of units mod n has order ϕ(n), and the result follows from (1.3.3).♣ We will often use the notation H ≤ G to indicate that H is a subgroup of G. If H is a proper subgroup, i.e. H ≤ G but H = G, we write H < G. 1.3.5 The Index is Multiplicative

If K ≤ H ≤ G then [G : K] = [G : H][H : K].

Proof. Choose representatives ai from each left coset of H in G, and representatives bj from each left coset of K in H. If cK is any left coset of K in G, then c ∈ ai H for some unique i, and if c = ai h, h ∈ H, then h ∈ bj K for some unique j, so that c belongs to ai bj K. The map (ai , bj ) → ai bj K is therefore onto, and it is one-to-one by the uniqueness of i and j. We therefore have a bijection between a set of size [G : H][H : K] and a set of size [G : K], as asserted. ♣ Now suppose that H and K are subgroups of G, and deﬁne HK to be the set of all products

page 8 of Chapter 1

hk, h ∈ H, k ∈ K. Note that HK need not be a group, since h1 k1 h2 k2 is not necessarily equal to h1 h2 k1 k2 . If G is abelian, then HK will be a group, and we have the following useful generalization of this observation. 1.3.6 Proposition If H ≤ G and K ≤ G, then HK ≤ G if and only if HK = KH. In this case, HK is the subgroup generated by H ∪ K. Proof. If HK is a subgroup, then (HK)−1 , the collection of all inverses of elements of HK, must coincide with HK. But (HK)−1 = K −1 H −1 = KH. Conversely, if HK = KH, then the inverse of an element in HK also belongs to HK, because (HK)−1 = K −1 H −1 = KH = HK. The product of two elements in HK belongs to HK, because (HK)(HK) = HKHK = HHKK = HK. The last statement follows from the observation that any subgroup containing H and K must contain HK. ♣ The set product HK deﬁned above suggests a multiplication operation on cosets. If H is a subgroup of G, we can multiply aH and bH, and it is natural to hope that we get abH. This does not always happen, but here is one possible criterion. 1.3.7 Lemma If H ≤ G, then (aH)(bH) = abH for all a, b ∈ G iﬀ cHc−1 = H for all c ∈ G. (Equivalently, cH = Hc for all c ∈ G.) Proof. If the second condition is satisﬁed, then (aH)(bH) = a(Hb)H = abHH = abH. Conversely, if the ﬁrst condition holds, then cHc−1 ⊆ cHc−1 H since 1 ∈ H, and (cH)(c−1 H) = cc−1 H(= H) by hypothesis. Thus cHc−1 ⊆ H, which implies that H ⊆ c−1 Hc. Since this holds for all c ∈ G, we have H ⊆ cHc−1 , and the result follows. ♣ Notice that we have proved that if cHc−1 ⊆ H for all c ∈ G, then in fact cHc−1 = H for all c ∈ G. 1.3.8 Deﬁnition Let H be a subgroup of G. If any of the following equivalent conditions holds, we say that H is normal subgroup of G, or that H is normal in G: 1. cHc−1 ⊆ H for all c ∈ G (equivalently, c−1 Hc ⊆ H for all c ∈ G) 2. cHc−1 = H for all c ∈ G (equivalently, c−1 Hc = H for all c ∈ G) 3. cH = Hc for all c ∈ G 4. Every left coset of H in G is also a right coset 5. Every right coset of H in G is also a left coset We have established the equivalence of 1,2 and 3 above, and 3 immediately implies 4. To show that 4 implies 3, suppose that cH = Hd. Then since c belongs to both cH and Hc, i.e., to both Hd and Hc, we must have Hd = Hc because right cosets partition G, so that any two right cosets must be either disjoint or identical. The equivalence of condition 5 is proved by a symmetrical argument. Notation: H G indicates that H is a normal subgroup of G; if H is a proper normal subgroup, we write H * G. 1.3.9 Deﬁnition of the Quotient Group If H is normal in G, we may deﬁne a group multiplication on cosets, as follows. If aH and bH are (left) cosets, let (aH)(bH) = abH; by (1.3.7), (aH)(bH) is simply the set product. If a1 is another member of aH and b1 another member of bH, then a1 H = aH and b1 H = bH (Problem 5). Therefore the set product of a1 H and b1 H is also abH. The point is that the product of two cosets does not depend on which representatives we select. To verify that cosets form a group under the above multiplication, we consider the four deﬁning requirements.

page 9 of Chapter 1

Closure: The product of two cosets is a coset. Associativity: This follows because multiplication in G is associative. Identity: The coset 1H = H serves as the identity. Inverse: The inverse of aH is a−1 H. The group of cosets of a normal subgroup N of G is called the quotient group of G by N ; it is denoted by G/N . Since the identity in G/N is 1N = N , we have, intuitively, “set everything in N equal to 1”. 1.3.10 Example Let GL(n, R) be the set of all nonsingular n by n matrices with real coeﬃcients, and let SL(n, r) be the subgroup formed by matrices whose determinant is 1 (GL stands for “general linear” and SL for “special linear”). Then SL(n, R) * GL(n, R), because if A is a nonsingular n by n matrix and B is n by n with determinant 1, then det(ABA−1 ) = (det A)(det B)(det A−1 ) = det B = 1. 1.3.11 Deﬁnition If f : G → H, where G and H are groups, then f is said to be a homomorphism if for all a, b in G, we have f (ab) = f (a)f (b). This idea will look familiar if G and H are abelian, in which case we write, using additive notation, f (a + b) = f (a) + f (b); thus a linear transformation on a vector space is, in particular, a homomorphism on the underlying abelian group. If f is a homomorphism from G to H, it must map the identity of G to the identity of H, since f (a) = f (a1G ) = f (a)f (1G ); multiply by f (a)−1 to get 1H = f (1G ). Furthermore, the inverse of f (a) is f (a−1 ), because 1 = f (aa−1 ) = f (a)f (a−1 ), so that [f (a)]−1 = f (a−1 ). 1.3.12 The Connection Between Homomorphisms and Normal Subgroups If f : G → H is a homomorphism, deﬁne the kernel of f as kerf = {a ∈ G : f (a) = 1}; then kerf is a normal subgroup of G. For if a ∈ G and b ∈ kerf , we must show that aba−1 belongs to kerf . But f (aba−1 ) = f (a)f (b)f (a−1 ) = f (a)(1)f (a)−1 = 1. Conversely, every normal subgroup is the kernel of a homomorphism. To see this, suppose that N G, and let H be the quotient group G/N . Deﬁne the map π : G → G/N by π(a) = aN ; π is called the natural or canonical map. Since π(ab) = abN = (aN )(bN ) = π(a)π(b), π is a homomorphism. The kernel of π is the set of all a ∈ G such that aN = N (= 1N ), or equivalently, a ∈ N . Thus ker π = N . 1.3.13 Proposition A homomorphism f is injective if and only if its kernel K is trivial, that is, consists only of the identity.

page 10 of Chapter 1

Proof. If f is injective and a ∈ K, then f (a) = 1 = f (1), hence a = 1. Conversely, if K is trivial and f (a) = f (b), then f (ab−1 ) = f (a)f (b−1 ) = f (a)[f (b)]−1 = f (a)[f (a)]−1 = 1, so ab−1 ∈ K. Thus ab−1 = 1, i.e., a = b, proving f injective. ♣ 1.3.14 Some Standard Terminology monomorphism = injective homomorphism epimorphism = surjective homomorphism isomorphism = bijective homomorphism endomorphism = homomorphism of a group to itself automorphism = isomorphism of a group with itself We close the section with a result that is often applied. 1.3.15 Proposition Let f : G → H be a homomorphism. (i) If K is a subgroup of G, then f (K) is a subgroup of H. If f is an epimorphism and K is normal, then f (K) is also normal. (ii) If K is a subgroup of H, then f −1 (K) is a subgroup of G. If K is normal, so is f −1 (K). Proof. (i) If f (a) and f (b) belong to f (K), so does f (a)f (b)−1 , since this element coincides with f (ab−1 ). If K is normal and c ∈ G, we have f (c)f (K)f (c)−1 = f (cKc−1 ) = f (K), so if f is surjective, then f (K) is normal. (ii) If a and b belong to f −1 (K), so does ab−1 , because f (ab−1 ) = f (a)f (b)−1 , which belongs to K. If c ∈ G and a ∈ f −1 (K) then f (cac−1 ) = f (c)f (a)f (c)−1 , so if K is normal, we have cac−1 ∈ f −1 (K), proving f −1 (K) normal. ♣ Problems For Section 1.3 In Problems 1-6, H is a subgroup of the group G, and a and b are elements of G. 1. Show that Ha = Hb iﬀ ab−1 ∈ H. 2. Show that “a ∼ b iﬀ ab−1 ∈ H” deﬁnes an equivalence relation. 3. If we deﬁne a and b to be equivalent iﬀ ab−1 ∈ H, show that the equivalence class of a is Ha. 4. Show that aH → Ha−1 is a one-to-one correspondence between left and right cosets of H. 5. If aH is a left coset of H in G and a1 ∈ aH, show that the left coset of H generated by a1 (i.e., a1 H), is also aH. 6. If [G : H] = 2, show that H is a normal subgroup of G. 7. Let S3 be the group of all permutations of {1, 2, 3}, and take a to be permutation (1,2,3), b the permutation (1,2), and e the identity permutation. Show that the elements of S3 are, explicitly, e , a , a2 , b , ab and a2 b. 8. Let H be the subgroup of S3 consisting of the identity e and the permutation b = (1, 2). Compute the left cosets and the right cosets of H in S3 . 9. Continuing Problem 8, show that H is not a normal subgroup of S3 . 10. Let f be an endomorphism of the integers Z. Show that f is completely determined by its action on 1. If f (1) = r, then f is multiplication by r, in other words, f (n) = rn for every integer n. 11. If f is an automorphism of Z, and I is the identity function on Z, show that f is either I or −I. 12. Since the composition of two automorphisms is an automorphism, and the inverse of an automorphism is an automorphism, it follows that the set of automorphisms of a group is a group under composition. In view of Problem 11, give a simple description of the group of automorphisms of Z. 13. Let H and K be subgroups of the group G. If x, y ∈ G, deﬁne x ∼ y iﬀ x can be written as hyk for some h ∈ H and k ∈ K. Show that ∼ is an equivalence relation.

page 11 of Chapter 1

14. The equivalence class of x ∈ G is HxK = {hxk : h ∈ H, k ∈ K}, which is called a double coset associated with the subgroups H and K. Thus the double cosets partition G. Show that any double coset can be written as a union of right cosets of H, or equally well as a union of left cosets of K. 1.4 The Isomorphism Theorems Suppose that N is a normal subgroup of G, f is a homomorphism from G to H, and π is the natural map from G to G/N , as pictured in Figure 1.4.1.

f H

G π

-

f

G/N Figure 1.4.1

We would like to ﬁnd a homomorphism f : G/N → H that makes the diagram commutative, that is, f (aN ) = f (a). Thus we get the same result by traveling directly from G to H via f as we do by going by the roundabout route via π followed by f . Here is the key result. 1.4.1 Factor Theorem Any homomorphism f whose kernel K contains N can be factored through G/N . In other words, in Figure 1.4.1 there is a unique homomorphism f : G/N → H such that f ◦ π = f . Furthermore, (i) f is an epimorphism if and only if f is an epimorphism; (ii) f is a monomorphism if and only if K = N ; (iii) f is an isomorphism if and only if f is an epimorphism and K = N . Proof. If the diagram is to commute, then f (aN ) must be f (a), and it follows that f , if it exists, is unique. The deﬁnition of f that we have just given makes sense, because if aN = bN , then a−1 b ∈ N ⊆ K, so f (a−1 b) = 1, and therefore f (a) = f (b). Since f (aN bN ) = f (abN ) = f (ab) = f (a)f (b) = f (aN )f (bN ), f is a homomorphism. By construction, f has the same image as f , proving (i). Now the kernel of f is {aN : f (a) = 1} = {aN : a ∈ K} = K/N. By (1.3.13), a homomorphism is injective, i.e., a monomorphism, if and only if its kernel is trivial. Thus f is a monomorphism if and only if K/N consists only of the identity element N . This means that if a is any element of K, then the coset aN coincides with N , which forces a to belong to N . Thus f is a monomorphism if and only if K = N , proving (ii). Finally, (iii) follows immediately from (i) and (ii). ♣ The factor theorem yields a fundamental result. 1.4.2 First Isomorphism Theorem If f : G → H is a homomorphism with kernel K, then the image of f is isomorphic to G/K. Proof. Apply the factor theorem with N = K, and note that f must be an epimorphism of G onto its image. ♣

page 12 of Chapter 1

If we are studying a subgroup K of a group G, or perhaps the quotient group G/K, we might try to construct a homomorphism f whose kernel is K and whose image H has desirable properties. The ﬁrst isomorphism theorem then gives G/K ∼ = H (where ∼ = is our symbol for isomorphism). If we know something about H, we may get some insight into K and G/K. We will prove several other isomorphism theorems after the following preliminary result. 1.4.3 Lemma Let H and N be subgroups of G, with N normal in G. Then (i) HN = N H, and therefore by (1.3.6), HN is a subgroup of G. (ii) N is a normal subgroup of HN . (iii) H ∩ N is a normal subgroup of H. Proof. (i) We have hN = N h for every h ∈ G, in particular for every h ∈ H. (ii) Since N is normal in G, it must be normal in the subgroup HN . (iii) H ∩ N is the kernel of the canonical map π : G → G/N , restricted to H. ♣ The subgroups we are discussing are related by a “parallelogram” or “diamond”, as Figure 1.4.2 suggests.

HN

H

N H ∩N Figure 1.4.2

1.4.4 Second Isomorphism Theorem If H and N are subgroups of G, with N normal in G, then H/(H ∩ N ) ∼ = HN/N. Note that we write HN/N rather than H/N , since N need not be a subgroup of H. Proof. Let π be the canonical epimorphism from G to G/N , and let π0 be the restriction of π to H. Then the kernel of π0 is H ∩ N , so by the ﬁrst isomorphism theorem, H/(H ∩ N ) is isomorphic to the image of π0 , which is {hN : h ∈ H} = HN/N . (To justify the last equality, note that for any n ∈ N we have hnN = hN.) ♣ 1.4.5 Third Isomorphism Theorem If N and H are normal subgroups of G, with N contained in H, then G/H ∼ = (G/N )/(H/N ), a “cancellation law”. Proof. This will follow directly from the ﬁrst isomorphism theorem if we can ﬁnd an epimorphism of G/N onto G/H with kernel H/N , and there is a natural candidate: f (aN ) = aH. To check that f is well-deﬁned, note that if aN = bN then a−1 b ∈ N ⊆ H, so aH = bH. Since a is an arbitrary element of G, f is surjective, and by deﬁnition of coset multiplication, f is a homomorphism. But the kernel of f is {aN : aH = H} = {aN : a ∈ H} = H/N. ♣ Now suppose that N is a normal subgroup of G. If H is a subgroup of G containing N , there is a natural analog of H in the quotient group G/N , namely the subgroup H/N . In fact we can make this correspondence very precise. Let ψ(H) = H/N

page 13 of Chapter 1

be a map from the set of subgroups of G containing N to the set of subgroups of G/N . We claim that ψ is a bijection. For if H1 /N = H2 /N then for any h1 ∈ H1 , we have h1 N = h2 N for some h2 ∈ H2 , so that h−1 2 h1 ∈ N , which is contained in H2 . Thus H1 ⊆ H2 , and by symmetry the reverse inclusion holds, so that H1 = H2 and ψ is injective. Now if Q is a subgroup of G/N and π : G → G/N is canonical, then π −1 (Q) = {a ∈ G : aN ∈ Q}, a subgroup of G containing N , and ψ(π −1 (Q)) = {aN : aN ∈ Q} = Q, proving ψ surjective. The map ψ has a number of other interesting properties, summarized in the following result, sometimes referred to as the fourth isomorphism theorem. 1.4.6 Correspondence Theorem If N is a normal subgroup of G, then the map ψ : H → H/N sets up a one-to-one correspondence between subgroups of G containing N and subgroups of G/N . The inverse of ψ is the map τ : Q → π −1 (Q), where π is the canonical epimorphism of G onto G/N . Furthermore, (i) H1 ≤ H2 if and only if H1 /N ≤ H2 /N , and in this case, [H2 : H1 ] = [H2 /N : H1 /N ] (ii) H is a normal subgroup of G if and only if H/N is a normal subgroup of G/N . More generally, (iii) H1 is a normal subgroup of H2 if and only if H1 /N is a normal subgroup of H2 /N , and in this case, H2 /H1 ∼ = (H2 /N )/H1 /N ). Proof. We have established that ψ is a bijection with inverse τ . If H1 ≤ H2 , we have H1 /N ≤ H2 /N immediately, and the converse follows from the above proof that ψ is injective. To prove the last statement of (i), let η map the left coset aH1 , a ∈ H2 , to the left coset (aN )(H1 /N ). Then η is a well-deﬁned and injective map of aH1 = bH1

iﬀ iﬀ iﬀ

a−1 b ∈ H1 (aN )−1 (bN ) = a−1 bN ∈ H1 /N (aN )(H1 /N ) = (bN )(H1 /N );

η is surjective because a ranges over all of H2 . To prove (ii), assume that H G; then for any a ∈ G we have (aN )(H/N )(aN )−1 = (aHa−1 )/N = H/N so that H/N G/N . Conversely, suppose that H/N is normal in G/N . Consider the homomorphism a → (aN )(H/N ), the composition of the canonical map of G onto G/N and the canonical map of G/N onto (G/N )/(H/N ). The element a will belong to the kernel of this map if and only if (aN )(H/N ) = H/N , which happens if and only if aN ∈ H/N , that is, aN = hN for some h ∈ H. But since N is contained in H, this statement is equivalent to a ∈ H. Thus H is the kernel of a homomorphism, and is therefore a normal subgroup of G. Finally, the proof of (ii) also establishes the ﬁrst part of (iii); just replace H by H1 and G by H2 . The second part of (iii) follows from the third isomorphism theorem (with the same replacement). ♣ We conclude the section with a useful technical result.

page 14 of Chapter 1

1.4.7 Proposition If H is a subgroup of G and N is a normal subgroup of G, we know by (1.4.3) that HN , the subgroup generated by H ∪ N , is a subgroup of G. If H is also a normal subgroup of G, then HN is normal in G as well. More generally, if for each i in the index set I, we have Hi G, then < Hi , i ∈ I >,the subgroup generated by the Hi (technically, by the set ∪i∈I Hi ) is a normal subgroup of G. Proof. A typical element in the subgroup generated by the Hi is a = a1 a2 · · · an where ak belongs to Hik . If g ∈ G then g(a1 a2 · · · an )g −1 = (ga1 g −1 )(ga2 g −1 ) · · · (gan g −1 ) and gak g −1 ∈ Hik because Hik G. Thus gag −1 belongs to < Hi , i ∈ I >. ♣ Problems For Section 1.4 1. Let Z be the integers, and nZ the set of integer multiples of n. Show that Z/nZ is isomorphic to Zn , the additive group of integers modulo n. (This is not quite a tautology if we view Zn concretely as the set {0, 1, . . . , n − 1}, with sums and diﬀerences reduced modulo n.) 2. If m divides n then Zm ≤ Zn ; for example, we can identify Z4 with the subgroup {0, 3, 6, 9} of Z12 . Show that Zn /Zm ∼ = Zn/m . 3. Let a be an element of the group G, and let fa : G → G be “conjugation by a”, that is, fa (x) = axa−1 , x ∈ G. Show that fa is an automorphism of G. 4. An inner automorphism of G is an automorphism of the form fa for some a ∈ G (see Problem 3). Show that the inner automorphisms of G form a group under composition of functions (a subgroup of the group of all automorphisms of G). 5. Let Z(G) be the center of G, that is, the set of all x in G such that xy = yx for all y in G. Thus Z(G) is the set of elements that commute with everything in G. Show that Z(G) is a normal subgroup of G, and that the group of inner automorphisms of G is isomorphic to G/Z(G). 6. If f is an automorphism of Zn , show that f is multiplication by m for some m relatively prime to n. Conclude that the group of automorphisms of Zn can be identiﬁed with the group of units mod n. 7. The diamond diagram associated with the second isomorphism theorem (1.4.4) illustrates least upper bounds and greatest lower bounds in a lattice. Verify that HN is the smallest subgroup of G containing both H and N , and H ∩ N is the largest subgroup of G contained in both H and N . 8. Let g be an automorphism of the group G, and fa an inner automorphism (see Problem 4). Show that g ◦fa ◦g −1 is an inner automorphism. Thus the group of inner automorphisms of G is a normal subgroup of the group of all automorphisms. 9. Identify a large class of groups for which the only inner automorphism is the identity mapping. 1.5 Direct Products 1.5.1 External and Internal Direct Products In this section we examine a popular construction. Starting with a given collection of groups, we build a new group with the aid of the cartesian product. Let’s start with two given groups H and K, and let G = H × K, the set of all ordered pairs (h, k), h ∈ H, k ∈ K. We deﬁne multiplication on G componentwise: (h1 , k1 )(h2 , k2 ) = (h1 h2 , k1 k2 ). Since (h1 h2 , k1 k2 ) belongs to G, it follows that G is closed under multiplication. The multiplication operation is associative because the individual products on H and K are

page 15 of Chapter 1

associative. The identity element in G is (1H , 1K ), and the inverse of (h, k) is (h−1 , k −1 ). Thus G is a group, called the external direct product of H and K. We may regard H and K as subgroups of G. More precisely, G contains isomorphic copies of H and K, namely H = {(h, 1K ) : h ∈ H} and K = {(1H , k) : k ∈ K.} Furthermore, H and K are normal subgroups of G. (Note that (h, k)(h1 , 1K )(h−1 , k −1 ) = (hh1 h−1 , 1K ), with hh1 h−1 ∈ H.) Also, from the deﬁnitions of H and K, we have G = H K and H ∩ K = {1}, where 1 = (1H , 1K ). If a group G contains normal subgroups H and K such that G = HK and H ∩K = {1}, we say that G is the internal direct product of H and K. Notice the key diﬀerence between external and internal direct products. We construct the external direct product from the component groups H and K. On the other hand, starting with a given group we discover subgroups H and K such that G is the internal direct product of H and K. Having said this, we must admit that in practice the distinction tends to be blurred, because of the following result. 1.5.2 Proposition If G is the internal direct product of H and K, then G is isomorphic to the external direct product H × K. Proof. Deﬁne f : H × K → G by f (h, k) = hk; we will show that f is an isomorphism. First note that if h ∈ H and k ∈ K then hk = kh. (Consider hkh−1 k −1 , which belongs to K since hkh−1 ∈ K, and also belongs to H since kh−1 k −1 ∈ H; thus hkh−1 k −1 = 1, so hk = kh.) (a) f is a homomorphism, since f ((h1 , k1 )(h2 , k2 )) = f (h1 h2 , k1 k2 ) = h1 h2 k1 k2 = (h1 k1 )(h2 k2 ) = f (h1 , k1 )f (h2 , k2 ).

(b) f is surjective, since by deﬁnition of internal direct product, G = HK. (c) f is injective, for if f (h, k) = 1 then hk = 1, so that h = k −1 .Thus h belongs to both H and K, so by deﬁnition of internal direct product, h is the identity, and consequently so is k. The kernel of f is therefore trivial. ♣ External and internal direct products may be deﬁned for any number of factors. We will restrict ourselves to a ﬁnite number of component groups, but the generalization to arbitrary cartesian products with componentwise multiplication is straightforward. 1.5.3 Deﬁnitions and Comments If H1 , H2 , . . . Hn are arbitrary groups, the external direct product of the Hi is the cartesian product G = H1 ×H2 ×· · ·×Hn , with componentwise multiplication: (h1 , h2 , ..., hn )(h1 , h2 , . . . hn ) = (h1 h1 , h2 h2 , . . . hn hn ); G contains an isomorphic copy of each Hi , namely H i = {(1H1 , . . . , 1Hi−1 , hi , 1Hi+1 , . . . , 1Hn ) : hi ∈ Hi }. As in the case of two factors, G = H 1 H 2 · · · H n , and H i G for all i; furthermore, if g ∈ G then g has a unique representation g = h1 h2 · · · hn where hi ∈ H i .

page 16 of Chapter 1

Speciﬁcally, g = (h1 , . . . , hn ) = (h1 , 1, . . . , 1) . . . (1, . . . , 1, hn ). The representation is unique because the only way to produce the ith component hi of g is for hi to be the ith component of the factor from H i . If a group G contains normal subgroups H1 , . . . , Hn such that G = H1 · · · Hn , and each g ∈ G can be uniquely represented as h1 · · · hn with hi ∈ Hi , i = 1, 2, . . . , n, we say that G is the internal direct product of the Hi . As in the case of two factors, if G is the internal direct product of the Hi , then G is isomorphic to the external direct product H1 × · · · × Hn ; the isomorphism f : H1 × · · · × Hn → G is given by f (h1 , . . . , hn ) = h1 · · · hn . The next result frequently allows us to recognize when a group is an internal direct product. 1.5.4 Proposition Suppose that G = H1 · · · Hn , where each Hi is a normal subgroup of G. The following conditions are equivalent: (1) G is the internal direct product of the Hi . (2) For all i = 1, 2, . . . , n, Hi

j=i

Hj = {1};

thus it does not matter in which order the Hi are listed. (3) For all i = 1, 2, . . . , n, Hi

i−1 j=1

Hj = {1}.

Proof. (1) implies (2): If g belongs to the product of the Hj , j = i, then g can be written as h1 · · · hn where hi = 1 and hj ∈ Hj for j = i. But if g also belongs to Hi then g can be written as k1 · · · kn where ki = g and kj = 1 for j = i. By uniqueness of representation in the internal direct product, hi = ki = 1 for all i, so g = 1. (2) implies (3): If g belongs to Hi and in addition, g = h1 · · · hi−1 with hj ∈ Hj , then g = h1 · · · hi−1 1Hi+1 · · · 1Hn , hence g = 1 by (2). (3) implies (1): If g ∈ G then since G = H1 · · · Hn we have g = h1 · · · hn with hi ∈ Hi . Suppose that we have another representation g = k1 · · · kn with ki ∈ Hi . Let i be the largest integer such that hi = ki . If i < n we can cancel the ht (= kt ), t > i, to get h1 · · · hi = k1 · · · ki . If i = n then h1 · · · hi = k1 · · · ki by assumption. Now any product of the Hi is a subgroup of G (as in (1.5.2), hi hj = hj hi for i = j, and the result follows from (1.3.6)). Therefore i−1 hi ki−1 ∈ Hj , j=1

and since ♣

hi ki−1

∈ Hi , we have

hi ki−1

= 1 by (3). Therefore hi = ki , which is a contradiction.

Problems For Section 1.5 1. Let C2 be a cyclic group of order 2, e.g., C2 = {1, a} where a2 = 1. Describe the multiplication table of the direct product C2 × C2 . Is C2 × C2 cyclic? 2. Show that C2 × C2 is isomorphic to the four group (Section 1.2, Problem 6). 3. Let Cn be a cyclic group of order n, e.g., Cn = {1, a, a2 , . . . an−1 } with an = 1. Show that the direct product C2 × C3 is cyclic, in particular it is isomorphic to C6 . 4. If n and m are relatively prime, show that Cn ×Cm is isomorphic to Cnm , and is therefore cyclic. 5. If n and m are not relatively prime, show that Cn × Cm is not cyclic. 6. If p and q are distinct primes and |G| = p, |H| = q, show that the direct product G × H is cyclic. 7. If H and K are arbitrary groups, show that H × K ∼ = K × H. 8. If G, H and K are arbitrary groups, show that G × (H × K) ∼ = (G × H) × K. In fact, both sides are isomorphic to G × H × K.

page 1 of Chapter 2

CHAPTER 2 RING FUNDAMENTALS 2.1 Basic Deﬁnitions and Properties 2.1.1 Deﬁnitions and Comments A ring R is an abelian group with a multiplication operation (a, b) → ab that is associative and satisﬁes the distributive laws: a(b + c) = ab + ac and (a + b)c = ab + ac for all a, b, c ∈ R. We will always assume that R has at least two elements, including a multiplicative identity 1R satisfying a1R = 1R a = a for all a in R. The multiplicative identity is often written simply as 1, and the additive identity as 0. If a, b, and c are arbitrary elements of R, the following properties are derived quickly from the deﬁnition of a ring; we sketch the technique in each case. (1) a0 = 0a = 0 [a0 + a0 = a(0 + 0) = a0; 0a + 0a = (0 + 0)a = 0a] (2) (−a)b = a(−b) = −(ab) [0 = 0b = (a+(−a))b = ab+(−a)b, so (−a)b = −(ab); similarly, 0 = a0 = a(b + (−b)) = ab + a(−b), so a(−b) = −(ab)] (3) (−1)(−1) = 1

[take a = 1, b = −1 in (2)]

(4) (−a)(−b) = ab [replace b by −b in (2)] (5) a(b − c) = ab − ac [a(b + (−c)) = ab + a(−c) = ab + (−(ac)) = ab − ac] (6) (a − b)c = ac − bc [(a + (−b))c = ac + (−b)c) = ac − (bc) = ac − bc] (7) 1 = 0 [If 1 = 0 then for all a we have a = a1 = a0 = 0, so R = {0}, contradicting the assumption that R has at least two elements] (8) The multiplicative identity is unique 1 = 11 = 1 ]

[If 1 is another multiplicative identity then

2.1.2 Deﬁnitions and Comments If a and b are nonzero but ab = 0, we say that a and b are zero divisors; if a ∈ R and for some b ∈ R we have ab = ba = 1, we say that a is a unit or that a is invertible. Note that ab need not equal ba; if this holds for all a, b ∈ R, we say that R is a commutative ring. An integral domain is a commutative ring with no zero divisors. A division ring or skew ﬁeld is a ring in which every nonzero element a has a multiplicative inverse a−1 (i.e., aa−1 = a−1 a = 1). Thus the nonzero elements form a group under multiplication. A ﬁeld is a commutative division ring. Intuitively, in a ring we can do addition, subtraction and multiplication without leaving the set, while in a ﬁeld (or skew ﬁeld) we can do division as well. Any ﬁnite integral domain is a ﬁeld. To see this, observe that if a = 0, the map x → ax, x ∈ R, is injective because R is an integral domain. If R is ﬁnite, the map is surjective as well, so that ax = 1 for some x. The characteristic of a ring R (written Char R) is the smallest positive integer such that n1 = 0, where n1 is an abbreviation for 1 + 1 + · · · 1 (n times). If n1 is never 0, we say that R has characteristic 0. Note that the characteristic can never be 1, since 1R = 0. If R is an integral domain and Char R = 0, then Char R must be a prime number. For if CharR = n = rs where r and s are positive integers greater than 1, then (r1)(s1) = n1 = 0, so either r1 or s1 is 0, contradicting the minimality of n. A subring of a ring R is a subset S or R that forms a ring under the operations of addition and multiplication deﬁned on R. In other words, S is an additive subgroup of R that contains 1R and is closed under multiplication. Note that 1R is automatically the multiplicative identity of S, since the multiplicative identity is unique (see (8) of (2.1.1)).

page 2 of Chapter 2

2.1.3 Examples 1. The integers Z form an integral domain that is not a ﬁeld. 2. Let Zn be the integers modulo n, that is, Zn = {0, 1, . . . , n−1} with addition and multiplication mod n. (If a ∈ Zn then a is identiﬁed with all integers a+kn, k = 0, ±1, ±2, . . .}.Thus, for example, in Z9 the multiplication of 3 by 4 results in 3 since 12 ≡ 3 mod 9, and therefore 12 is identiﬁed with 3. Zn is a ring, which is an integral domain (and therefore a ﬁeld, since Zn is ﬁnite) if and only if n is prime. For if n = rs then rs = 0 in Zn ; if n is prime then every nonzero element in Zn has a multiplicative inverse, by Fermat’s little theorem 1.3.4. Note that by deﬁnition of characteristic, any ﬁeld of prime characteristic p contains an isomorphic copy of Zp . Any ﬁeld of characteristic 0 contains a copy of Z , hence a copy of the rationals Q. 3. If n ≥ 2, then the set Mn (R) of all n by n matrices with coeﬃcients in a ring R forms a noncommutative ring, with the identity matrix In as multiplicative identity. If we identify the element c ∈ R with the diagonal matrix cIn , we may regard R as a subring of Mn (R). It is possible for the product of two nonzero matrices to be zero, so that Mn (R) is not an integral domain. (To generate a large class of examples, let Eij be the matrix with 1 in row i, column j, and 0’s elsewhere. Then Eij Ekl = δjk Eil , where δjk is 1 when j = k, and 0 otherwise.) 4. Let 1, i, j and k be basis vectors in 4-dimensional Euclidean space, and deﬁne multiplication of these vectors by i2 = j 2 = k 2 = −1, ij = k, jk = i, ki = j, ji = −ij, kj = −jk, ik = −ki

(1)

Let H be the set of all linear combinations a + bi + cj + dk where a, b, c and d are real numbers. Elements of H are added componentwise and multiplied according to the above rules, i.e., (a + bi + cj + dk)(x + yi + zj + wk) = (ax − by − cz − dw) + (ay + bx + cw − dz)i +(az + cx + dy − bw)j + (aw + dx + bz − cy)k. H (after Hamilton) is called the ring of quaternions. In fact H is a division ring; the inverse of a + bi + cj + dk is (a2 + b2 + c2 + d2 )−1 (a − bi − cj − dk). H can also be represented by 2 by 2 matrices with complex entries, with multiplication of quaternions corresponding to ordinary matrix multiplication. To see this, let 1=

1 0

0 i , i= 1 0

0 0 , j= −i −1

1 0 , k= 0 i

i ; 0

a direct computation shows that 1, i, j and k obey the multiplication rules (1) given above. Thus we may identify the quaternion a + bi + cj + dk with the matrix a1 + bi + cj + dk = (where in the matrix, i is

√

a + bi −c + di

c + di a − bi

−1, not the quaternion i).

The set of 8 elements ±1, ±i, ±j, ±k forms a group under multiplication; it is called the quaternion group.

page 3 of Chapter 2

5. If R is a ring, then R[X], the set of all polynomials in X with coeﬃcients in R, is also a ring under ordinary polynomial addition and multiplication, as is R[X1 , . . . , Xn ], the set of polynomials in n variables Xi , 1 ≤ i ≤ n, with coeﬃcients in R. Formally, the polynomial A(X) = a0 + a1 X + · · · + an X n is simply the sequence (a0 , . . . , an ); the symbol X is a placeholder. The product of two polynomials A(X) and B(X) is a polynomial whose X k coeﬃcient is a0 bk + a1 bk−1 + · · · + ak b0 . If we wish to evaluate a polynomial on R, we use the evaluation map a0 + a1 X + · · · + an X n → a0 + a1 x + · · · + an xn where x is a particular element of R. A nonzero polynomial can evaluate to 0 at all points of R. For example, X 2 + X evaluates to 0 on Z2 , the ﬁeld of integers modulo 2, since 1 + 1 = 0 mod 2. We will say more about evaluation maps in Section 2.5, when we study polynomial rings. 6. If R is a ring, then R[[X]], the set of formal power series a0 + a1 X + a2 X 2 + · · · with coeﬃcients in R, is also a ring under ordinary addition and multiplication of power series. The deﬁnition of multiplication is purely formal and convergence is never mentioned; we simply deﬁne the coeﬃcient of X n in the product of a0 + a1 X + a2 X 2 + · · · and b0 + b1 X + b2 X 2 + · · · to be a0 bn + a1 bn−1 + · · · + an−1 b1 + an b0 . In Examples 5 and 6, if R is an integral domain, so are R[X] and R[[X]]. In Example 5, look at leading coeﬃcients to show that if f (X) = 0 and g(X) = 0, then f (X)g(X) = 0. In Example 6, if f (X)g(X) = 0 with f (X) = 0, let ai be the ﬁrst nonzero coeﬃcient of f (X). Then ai bj = 0 for all j, and therefore g(X) = 0. 2.1.4 Lemma The generalized associative law holds for multiplication in a ring. There is also a generalized distributive law: (a1 + · · · + am )(b1 + · · · + bn ) =

m n

ai bj .

i=1 j=1

Proof. The argument for the generalized associative law is exactly the same as for groups; see the beginning of Section 1.1. The generalized distributive law is proved in two stages. First set m = 1 and work by induction on n, using the left distributive law a(b + c) = ab + ac. Then use induction on m and the right distributive law (a + b)c = ac + bc on (a1 + · · · + am + am+1 )(b1 + · · · + bn ). ♣ n 2.1.5 Proposition The Binomial Theorem (a + b)n = k=0 nk ak bn−k is valid in any ring, if ab = ba. Proof. The standard proof via elementary combinatorial analysis works. Speciﬁcally, (a + b)n = (a + b) · · · (a + b), and we can expand this product by multiplying an element (a or b) from object 1 (the ﬁrst (a + b)) times an element from object 2 times · · ·times an element from object n, in all possible ways. Since ab = ba, these terms are of the form ak bn−k , 0 ≤ k ≤ n. The number of terms corresponding to a given k is the number of ways of selecting k objects from a collection of n, namely nk . ♣ Problems For Section 2.1 1. If R is a ﬁeld, is R[X] a ﬁeld always? sometimes? never? 2. If R is a ﬁeld, what are the units of R[X]? 3. Consider the ring of formal power series with rational coeﬃcients.

page 4 of Chapter 2

(a) Give an example of a nonzero element that does not have a multiplicative inverse, and thus is not a unit. (b) Give an example of a nonconstant element (one that is not simply a rational number) that does have a multiplicative inverse, and therefore is a unit. √ 4. Let Z[i] be the ring of Gaussian integers a + bi, where i = −1 and a and b are integers. Show that Z[i] is an integral domain that is not a ﬁeld. 5. What are the units of Z[i]? 6. Establish the following quaternion identities:

(a)

(x1 + y1 i + z1 j + w1 k)(x2 − y2 i − z2 j − w2 k) = (x1 x2 + y1 y2 + z1 z2 + w1 w2 ) +(−x1 y2 + y1 x2 − z1 w2 + w1 z2 )i + (−x1 z2 + z1 x2 − w1 y2 + y1 w2 )j +(−x1 w2 + w1 x2 − y1 z2 + z1 y2 )k

(b)

(x2 + y2 i + z2 j + w2 k)(x1 − y1 i − z1 j − w1 k) = (x1 x2 + y1 y2 + z1 z2 + w1 w2 ) +(x1 y2 − y1 x2 + z1 w2 − w1 z2 )i + (x1 z2 − z1 x2 + w1 y2 − y1 w2 )j +(x1 w2 − w1 x2 + y1 z2 − z1 y2 )k

(c) The product of a quaternion h = a + bi + cj + dk and its conjugate h∗ = a − bi − cj − dk is a2 + b2 + c2 + d2 . If q and t are quaternions, then (qt)∗ = t∗ q ∗ . 7. Use Problem 6 to establish Euler’s Identity for real numbers xr , yr , zr , wr , r = 1, 2:

(x21 + y12 + z12 + w12 )(x22 + y22 + z22 + w22 ) = (x1 x2 + y1 y2 + z1 z2 + w1 w2 )2 +(x1 y2 − y1 x2 + z1 w2 − w1 z2 )2 + (x1 z2 − z1 x2 + w1 y2 − y1 w2 )2 +(x1 w2 − w1 x2 + y1 z2 − z1 y2 )2 8. Recall that an endomorphism of a group G is a homomorphism of G to itself. Thus if G is abelian, an endomorphism is a function f : G → G such that f (a + b) = f (a) + f (b) for all a, b ∈ G. Deﬁne addition of endomorphisms in the natural way:(f + g)(a) = f (a) + g(a), and deﬁne multiplication as functional composition:(f g)(a) = f (g(a)). Show that the set End G of endomorphisms of G becomes a ring under these operations. 9. What are the units of End G? 10. It can be shown that every positive integer is the sum of 4 squares. A key step is to prove that if n and m can be expressed as sums of 4 squares, so can nm. Do this using Euler’s identity, and illustrate for the case n = 34, m = 54. 11. Which of the following collections of n by n matrices form a ring under matrix addition and multiplication? (a) symmetric matrices (b) matrices whose entries are 0 except possibly in column 1 (c) lower triangular matrices (aij = 0 for i < j) (d) upper triangular matrices (aij = 0 for i > j) 2.2 Ideals, Homomorphisms, and Quotient Rings Let f : R → S, where R and S are rings. Rings are, in particular, abelian groups under addition, so we know what it means for f to be a group homomorphism: f (a+b) = f (a)+f (b)

page 5 of Chapter 2

for all a, b in R. It is then automatic that f (0R ) = 0S (see (1.3.11)). It is natural to consider mappings f that preserve multiplication as well as addition, i.e., f (a + b) = f (a) + f (b) and f (ab) = f (a)f (b) for all a, b ∈ R. But here it does not follow that f maps the multiplicative identity 1R to the multiplicative identity 1S . We have f (a) = f (a1R ) = f (a)f (1R ), but we cannot multiply on the left by f (a)−1 , which might not exist. We avoid this diﬃculty by only considering functions f that have the desired behavior. 2.2.1 Deﬁnition If f : R → S, where R and S are rings, we say that f is a ring homomorphism if f (a + b) = f (a) + f (b) and f (ab) = f (a)f (b) for all a, b ∈ R, and f (1R ) = 1S . 2.2.2 Example Let f : Z → Mn (R), n ≥ 2, be deﬁned by f (n) = nE11 (see (2.1.3), example 3). Then we have f (a + b) = f (a) + f (b), f (ab) = f (a)f (b), but f (1) = In . Thus f is not a ring homomorphism. In Chapter 1, we proved the basic isomorphism theorems for groups, and a key observation was the connection between group homomorphisms and normal subgroups. We can prove similar theorems for rings, but ﬁrst we must replace the normal subgroup by an object that depends on multiplication as well as addition. 2.2.3 Deﬁnitions and Comments Let I be a subset of the ring R, and consider the following three properties: (1) I is an additive subgroup of R (2) If a ∈ I and r ∈ R then ra ∈ I, in other words, rI ⊆ I for every r ∈ R (3) If a ∈ I and r ∈ R then ar ∈ I, in other words, Ir ⊆ I for every r ∈ R If (1) and (2) hold, I is said to be a left ideal of R. If (1) and (3) hold, I is said to be a right ideal of R. If all three properties are satisﬁed, I is said to be an ideal (or two-sided ideal) of R, a proper ideal if I = R, a nontrivial ideal if I is neither R nor {0}. If f : R → S is a ring homomorphism, its kernel is kerf = {r ∈ R : f (r) = 0}; exactly as in (1.3.13), f is injective if and only if kerf = {0}. Now it follows from the deﬁnition of ring homomorphism that kerf is an ideal of R. The kernel must be a proper ideal because if kerf = R then f is identically 0, in particular, f (1R ) = 1S = 0S , a contradiction (see (7) of (2.1.1)). Conversely, every proper ideal is the kernel of a ring homomorphism, as we will see in the discussion to follow. 2.2.4 Construction of Quotient Rings Let I be a proper ideal of the ring R. Since I is a subgroup of the additive group of R, we can form the quotient group R/I, consisting of cosets r + I, r ∈ R. We deﬁne multiplication of cosets in the natural way: (r + I)(s + I) = rs + I. To show that multiplication is well-deﬁned, suppose that r + I = r + I and s + I = s + I, so that r − r is an element of I, call it a, and s − s is an element of I, call it b. Thus r s = (r + a)(s + b) = rs + as + rb + ab, and since I is an ideal, we have as ∈ I, rb ∈ I, and ab ∈ I. Consequently, r s + I = rs + I, so the multiplication of two cosets is independent of the particular representatives r and s that we choose.

page 6 of Chapter 2

From our previous discussion of quotient groups, we know that the cosets of the ideal I from a group under addition, and the group is abelian because R itself is an abelian group under addition. Since multiplication of cosets r + I and s + I is accomplished simply by multiplying the coset representatives r and s in R and then forming the coset rs + I, we can use the ring properties of R to show that the cosets of I form a ring, called the quotient ring of R by I. The identity element of the quotient ring is 1R +I, and the zero element is 0R +I. Furthermore, if R is a commutative ring, so is R/I. The fact that I is proper is used in verifying that R/I has at least two element. For if 1R + I = 0R + I, then 1R = 1R − 0R ∈ I; thus for any r ∈ R we have r = r1R ∈ I, so that R = I, a contradiction. 2.2.5 Proposition Every proper ideal I is the kernel of a ring homomorphism. Proof. Deﬁne the natural or canonical map π : R → R/I by π(r) = r + I. We already know that π is a homomorphism of abelian groups and its kernel is I (see (1.3.12)). To verify that π preserves multiplication, note that π(rs) = rs + I = (r + I)(s + I) = π(r)π(s); since π(1R ) = 1R + I = 1R/I , π is a ring homomorphism. ♣ 2.2.6 Proposition If f : R → S is a ring homomorphism and the only ideals of R are {0} and R, then f is injective. (In particular, if R is a division ring, then R satisﬁes this hypothesis.) Proof. Let I = kerf , an ideal of R (see (2.2.3)). If I = R then f is identically zero, and is therefore not a legal ring homomorphism since f (1R ) = 1S = 0S . Thus I = {0}, so that f is injective. If R is a division ring, then in fact R has no nontrivial left or right ideals. For suppose that I is a left ideal of R and a ∈ I, a = 0. Since R is a division ring, there is an element b ∈ R such that ba = 1, and since I is a left ideal, we have 1 ∈ I, which implies that I = R. If I is a right ideal, we choose the element b such that ab = 1. ♣ 2.2.7 Deﬁnitions and Comments If X is a nonempty subset of the ring R, then < X > will denote the ideal generated by X, that is, the smallest ideal of R that contains X.Explicitly, < X > = RXR = the collection of ﬁnite sums of the form

ri xi si with ri , si ∈ R and xi ∈ X.

i

To show that this is correct, verify that the ﬁnite sums of the given type form an ideal containing X. On the other hand, if J is any ideal containing X, then all ﬁnite sums i ri xi si must belong to J. If R is commutative, then rxs = rsx, and we may as well drop the s. In other words: In a commutative ring , < X >= RX = all ﬁnite sums

ri xi , ri ∈ R, xi ∈ X.

i

An ideal generated by a single element a is called a principal ideal and is denoted by < a > or (a). In this case, X = {a}, and therefore: In a commutative ring , the principal ideal generated by a is < a >= {ra : r ∈ R}, the set of all multiples of a, sometimes denoted by Ra.

page 7 of Chapter 2

2.2.8 Deﬁnitions and Comments In an arbitrary ring, we will sometimes need to consider the sum of two ideals I and J, deﬁned as {x + y : x ∈ I, y ∈ J}. It follows from the distributive laws that I + J is also an ideal. Similarly, the sum of two left [resp. right] ideals is a left [resp. right] ideal. Problems For Section 2.2 1. What are the ideals in the ring of integers? 2. Let Mn (R) be the ring of n by n matrices with coeﬃcients in the ring R. If Ck is the subset of Mn (R) consisting of matrices that are 0 except perhaps in column k, show that Ck is a left ideal of Mn (R). Similarly, if Rk consists of matrices that are 0 except perhaps in row k, then Rk is a right ideal of Mn (R). 3. In Problem 2, assume that R is a division ring, and let Eij be the matrix with 1 in row i, column j, and 0’s elsewhere. (a) If A ∈ Mn (R), show that Eij A has row j of A as its ith row, with 0’s elsewhere. (b) Now suppose that A ∈ Ck . Show that Eij A has ajk in the ik position, with 0’s elsewhere, so that if ajk is not zero, then a−1 jk Eij A = Eik . (c) If A is a nonzero matrix in Ck with ajk = 0, and C is any matrix in Ck , show that n

cik a−1 jk Eij A = C.

i=1

4. Continuing Problem 3, if a nonzero matrix A in Ck belongs to the left ideal I of Mn (R), show that every matrix in Ck belongs to I. Similarly, if a nonzero matrix A in Rk belongs to the right ideal I of Mn (R), every matrix in Rk belongs to I. 5. Show that if R is a division ring, then Mn (R) has no nontrivial two-sided ideals. 6. In R[X], express the set I of polynomials with no constant term as < f > for an appropriate f , and thus show that I is a principal ideal. 7. Let R be a commutative ring whose only proper ideals are {0} and R. Show that R is a ﬁeld. 8. Let R be the ring Zn of integers modulo n, where n may be prime or composite. Show that every ideal of R is principal. 2.3 The Isomorphism Theorems For Rings The basic ring isomorphism theorems may be proved by adapting the arguments used in Section 1.4 to prove the analogous theorems for groups. Suppose that I is an ideal of the ring R, f is a ring homomorphism from R to S with kernel K, and π is the natural map, as indicated in Figure 2.3.1. To avoid awkward analysis of special cases, let us make a blanket assumption that any time a quotient ring R0 /I0 appears in the statement of a theorem, the ideal I0 is proper.

f S

R π -

f R/I

Figure 2.3.1 2.3.1 Factor Theorem For Rings Any ring homomorphism whose kernel contains I can be factored through R/I. In other words, in Figure 2.3.1 there is a unique ring homomorphism f : R → S that makes the diagram commutative. Furthermore,

page 8 of Chapter 2

(i) f is an epimorphism if and only if f is an epimorphism; (ii) f is a monomorphism if and only if kerf = I; (iii) f is an isomorphism if and only if f is an epimorphism and kerf = I. Proof. The only possible way to deﬁne f is f (a + I) = f (a). To verify that f is well-deﬁned, note that if a + I = b + I, then a − b ∈ I ⊆ K, so f (a − b) = 0, i.e.,f (a) = f (b). Since f is a ring homomorphism, so is f . To prove (i),(ii) and (iii), the discussion in (1.4.1) may be translated into additive notation and copied. ♣ 2.3.2 First Isomorphism Theorem For Rings If f : R → S is a ring homomorphism with kernel K, then the image of f is isomorphic to R/K. Proof. Apply the factor theorem with I = K, and note that f is an epimorphism onto its image. ♣ 2.3.3 Second Isomorphism Theorem For Rings Let I be an ideal of the ring R, and let S be a subring of R. Then (a) S + I(= {x + y : x ∈ S, y ∈ I}) is a subring of R; (b) I is an ideal of S + I; (c) S ∩ I is an ideal of S; (d) (S + I)/I is isomorphic to S/(S ∩ I), as suggested by the “parallelogram” or“diamond” diagram in Figure 2.3.2.

S + I

S

I S ∩I Figure 2.3.2

Proof. (a) Verify directly that S + I is an additive subgroup of R that contains 1R (since 1R ∈ S and 0R ∈ I) and is closed under multiplication. For example, if a ∈ S, x ∈ I, b ∈ S, y ∈ I, then (a + x)(b + y) = ab + (ay + xb + xy) ∈ S + I. (b) Since I is an ideal of R, it must be an ideal of the subring S + I. (c) This follows from the deﬁnitions of subring and ideal. (d) Let π : R → R/I be the natural map, and let π0 be the restriction of π to S. Then π0 is a ring homomorphism whose kernel is S ∩ I and whose image is {a + I : a ∈ S} = (S + I)/I. (To justify the last equality, note that if s ∈ S and x ∈ I we have (s + x) + I = s + I.) By the ﬁrst isomorphism theorem for rings, S/ker π0 is isomorphic to the image of π0 , and (d) follows. ♣ 2.3.4 Third Isomorphism Theorem For Rings Let I and J be ideals of the ring R, with I ⊆ J. Then J/I is an ideal of R/I, and R/J ∼ = (R/I)/(J/I). Proof. Deﬁne f : R/I → R/J by f (a + I) = a + J. To check that f is well-deﬁned, suppose that a + I = b + I. Then a − b ∈ I ⊆ J, so a + J = b + J. By deﬁnition of addition and multiplication of cosets in a quotient ring, f is a ring homomorphism. Now kerf = {a + I : a + J = J} = {a + I : a ∈ J} = J/I and imf = {a + J : a ∈ R} = R/J.

page 9 of Chapter 2

(where im denotes image). The result now follows from the ﬁrst isomorphism theorem for rings. ♣ 2.3.5 Correspondence Theorem For Rings If I is an ideal of the ring R, then the map S → S/I sets up a one-to-one correspondence between the set of all subrings of R containing I and the set of all subrings of R/I, as well as a one-to-one correspondence between the set of all ideals of R containing I and the set of all ideals of R/I. The inverse of the map is Q → π −1 (Q), where π is the canonical map: R → R/I. Proof. The correspondence theorem for groups yields a one-to-one correspondence between additive subgroups or R containing I and additive subgroups of R/I. We must check that subrings correspond to subrings and ideals to ideals. If S is a subring of R then S/I is closed under addition, subtraction and multiplication. For example, if s and s belong to S, we have (s + I)(s + I) = ss + I ∈ S/I. Since 1R ∈ S we have 1R + I ∈ S/I, proving that S/I is a subring of R/I. Conversely, if S/I is a subring of R/I, then S is closed under addition, subtraction and multiplication, and contains the identity, hence is a subring or R. For example, if s, s ∈ S then (s + I)(s + I) ∈ S/I, so that ss + I = t + I for some t ∈ S, and therefore ss − t ∈ I. But I ⊆ S, so ss ∈ S. Now if J is an ideal of R containing I, then J/I is an ideal of R/I by the third isomorphism theorem for rings. Conversely, let J/I be an ideal of R/I. If r ∈ R and x ∈ J then (r +I)(x+I) ∈ J/I, that is, rx+I ∈ J/I. Thus for some j ∈ J we have rx−j ∈ I ⊆ J, so rx ∈ J. A similar argument shows that xr ∈ J, and that J is an additive subgroup of R. It follows that J is an ideal of R. ♣ We now consider the Chinese remainder theorem, which is an abstract version of a result in elementary number theory. Along the way, we will see a typical application of the ﬁrst isomorphism theorem for rings, and in fact the development of any major theorem of algebra is likely to include an appeal to one or more of the isomorphism theorems. The following observations may make the ideas easier to visualize. 2.3.6 Deﬁnitions and Comments (i) If a and b are integers that are congruent modulo n, then a − b is a multiple of n. Thus a − b belongs to the ideal In consisting of all multiples of n in the ring Z of integers. Thus we may say that a is congruent to b modulo In . In general, if a, b ∈ R and I is an ideal of R, we say that a ≡ b mod I if a − b ∈ I. (ii) The integers a and b are relatively prime if and only if the integer 1 can be expressed as a linear combination of a and b. Equivalently, the sum of the ideals Ia and Ib is the entire ring Z. In general, we say that the ideals I and J in the ring R are relatively prime if I + J = R. (iii) If Ini consists of all multiples of ni in the ring of integers (i = 1, . . . k), then the intersection ∩ki=1 Ini is Ir , where r is the least common multiple of the ni . If the ni are relatively prime in pairs, then r is the product of the ni . (iv) If R1 , . . . , Rn are rings, the direct product of the Ri is deﬁned as the ring of n-tuples (a1 , . . . , an ), ai ∈ Ri , with componentwise addition and multiplication, that is, (a1 , . . . , an ) + (b1 , . . . , bn ) = (a1 + b1 , . . . , an + bn ) and (a1 , . . . , an )(b1 , . . . , bn ) = (a1 b1 , . . . , an bn ). The zero element is (0, . . . , 0) and the multiplicative identity is (1, . . . , 1). 2.3.7 Chinese Remainder Theorem Let R be an arbitrary ring, and let I1 , . . . , In be ideals in R that are relatively prime in pairs, that is, Ii + Ij = R for all i = j.

page 10 of Chapter 2

(1) If a1 = 1 (the multiplicative identity of R) and aj = 0 (the zero element of R) for j = 2, . . . , n, then there is an element a ∈ R such that a ≡ ai mod Ii for all i = 1, . . . , n. More generally, (2) If a1 , . . . , an are arbitrary elements of R, there is an element a ∈ R such that a ≡ ai mod Ii for all i = 1, . . . , n. (3) If b is another element of R such that b ≡ ai mod Ii for all i = 1, . . . , n, then b ≡ a mod I1 ∩ I2 ∩ . . . ∩ In . Conversely, if b ≡ a mod ∩ni=1 Ii , then b ≡ ai mod Ii for all i. n n (4) R/ i=1 Ii is isomorphic to the direct product i=1 R/Ii . Proof. (1) If j > 1 we have I1 + Ij = R, so there exist elements bj ∈ I1 and cj ∈ Ij such that bj + cj = 1; thus n

(bj + cj ) = 1. j=2

Expand the left side and observe that any product containing at least one bj belongs to I1 , n while c2 · · · cn belongs to j=2 Ij , the collection of all ﬁnite sums of products x2 · · · xn with n xj ∈ Ij . Thus we have elements b ∈ I1 and a ∈ j=2 Ij (a subset of each Ij ) with b + a = 1. Consequently, a ≡ 1 mod I1 and a ≡ 0 mod Ij for j > 1, as desired. (2) By the argument of part (1), for each i we can ﬁnd ci with ci ≡ 1 mod Ii and ci ≡ 0 mod Ij , j = i. If a = a1 c1 + · · · + an cn , then a has the desired properties. To see this, write a − ai = a − ai ci + ai (ci − 1), and note that a − ai ci is the sum of the aj cj , j = i, and is therefore congruent to 0 mod Ii . (3) We have b ≡ ai mod Ii for all i iﬀ b − a ≡ 0 mod Ii for all i, that is, iﬀ b − a ∈ ∩ni=1 Ii , and the result follows. n (4) Deﬁne f : R → i=1 R/Ii by f (a) = (a + I1 , . . . , a + In ). If a1 , . . . , an ∈ R, then by part (2) there is an element a ∈ R such that a ≡ ai mod Ii for all i. But then f (a) = (a1 + I1 , . . . , an + In ), proving that f is surjective. Since the kernel of f is the intersection of the ideals Ij , the result follows from the ﬁrst isomorphism theorem for rings. ♣ The concrete version of the Chinese remainder theorem can be recovered from the abstract result; see Problems 3 and 4. Problems For Section 2.3 1. Show that the group isomorphisms of Section 1.4, Problems 1 and 2, are ring isomorphisms as well. 2. Give an example of an ideal that is not a subring, and a subring that is not an ideal. 3. If the integers mi , i = 1, . . . , n, are relatively prime in pairs, and a1 , . . . , an are arbitrary integers, show that there is an integer a such that a ≡ ai mod mi for all i, and that any two such integers are congruent modulo m1 · · · mn . 4. If the integers mi , i = 1, . . . , n, are relatively prime in pairs and m = m1 · · · mn , show n that there is a ring isomorphism between Zm and the direct product i=1 Zmi . Speciﬁcally, a mod m corresponds to (a mod m1 , . . . , a mod mn ). 5. Suppose that R = R1 × R2 is a direct product of rings. Let R1 be the ideal R1 × {0} = {(r1 , 0) : r1 ∈ R1 }, and let R2 be the ideal {(0, r2 ) : r2 ∈ R2 }. Show that R/R1 ∼ = R2 and R/R2 ∼ = R1 . If I1 , . . . , In are ideals, the product I1 · · · In is deﬁned as the set of all ﬁnite sums i a1i a2i · · · ani , where aki ∈ Ik , k = 1, . . . , n. [See the proof of part (1) of (2.3.7); check that the product of ideals is an ideal.] 6. Under the hypothesis of the Chinese remainder theorem, show that the intersection of the ideals Ii coincides with their product.

page 11 of Chapter 2

7. Let I1 , . . . , In be ideals in the ring R. Suppose that R/ ∩i Ii is isomorphic to i R/Ii via a + ∩i Ii → (a + I1 , . . . , a + In ). Show that the ideals Ii are relatively prime in pairs. 2.4 Maximal and Prime Ideals If I is an ideal of the ring R, we might ask“What is the smallest ideal containing I?” and “What is the largest ideal containing I?” Neither of these questions is challenging; the smallest ideal is I itself, and the largest ideal is R. But if I is a proper ideal and we ask for a maximal proper ideal containing I, the question is much more interesting. 2.4.1 Deﬁnition A maximal ideal in the ring R is a proper ideal that is not contained in any strictly larger proper ideal. 2.4.2 Theorem Every proper ideal I of the ring R is contained in a maximal ideal. Consequently, every ring has at least one maximal ideal. Proof. The argument is a prototypical application of Zorn’s lemma. Consider the collection of all proper ideals containing I, partially ordered by inclusion. Every chain {Jt , t ∈ T } of proper ideals containing I has an upper bound, namely the union of the chain. (Note that the union is still a proper ideal, because the identity 1R belongs to none of the ideals Jt .) By Zorn, there is a maximal element in the collection, that is, a maximal ideal containing I. Now take I = {0} to conclude that every ring has at least one maximal ideal. ♣ We have the following characterization of maximal ideals. 2.4.3 Theorem Let M be an ideal in the commutative ring R. Then M is a maximal ideal if and only if R/M is a ﬁeld. Proof. Suppose M is maximal. We know that R/M is a ring (see (2.2.4)); we need to ﬁnd the multiplicative inverse of the element a+M of R/M , where a+M is not the zero element, i.e., a ∈ / M . Since M is maximal, the ideal Ra+M , which contains a and is therefore strictly larger than M , must be the ring R itself. Consequently, the identity element 1 belongs to Ra + M . If 1 = ra + m where r ∈ R and m ∈ M , then (r + M )(a + M ) = ra + M = (1 − m) + M = 1 + M since m ∈ M, proving that r + M is the multiplicative inverse of a + M . Conversely, if R/M is a ﬁeld, then M must be a proper ideal. If not, then M = R, so that R/M contains only one element, contradicting the requirement that 1 = 0 in R/M (see (7) of (2.1.1)). By (2.2.6), the only ideals of R/M are {0} and R/M , so by the correspondence theorem (2.3.5), there are no ideals properly between M and R. Therefore M is a maximal ideal. ♣ If in (2.4.3) we relax the requirement that R/M be a ﬁeld, we can identify another class of ideals. 2.4.4 Deﬁnition A prime ideal in a commutative ring R is a proper ideal P such that for any two elements a, b in R, ab ∈ P implies that a ∈ P or b ∈ P. We can motivate the deﬁnition by looking at the ideal (p) in the ring of integers. In this case, a ∈ (p) means that p divides a, so that (p) will be a prime ideal if and only if p divides ab implies that p divides a or p divides b, which is equivalent to the requirement that p be a prime number.

page 12 of Chapter 2

2.4.5 Theorem If P is an ideal in the commutative ring R, then P is a prime ideal if and only if R/P is an integral domain. Proof. Suppose P is prime. Since P is a proper ideal, R/P is a ring. We must show that if (a + P )(b + P ) is the zero element P in R/P , then a + P = P or b + P = P , i.e., a ∈ P or b ∈ P . This is precisely the deﬁnition of a prime ideal. Conversely, if R/P is an integral domain, then as in (2.4.3), P is a proper ideal. If ab ∈ P , then (a + P )(b + P ) is zero in R/P , so that a + P = P or b + P = P , i.e., a ∈ P or b ∈ P . ♣ 2.4.6 Corollary In a commutative ring, a maximal ideal is prime. Proof. This is immediate from (2.4.3) and (2.4.5). ♣ 2.4.7 Corollary Let f : R → S be an epimorphism of commutative rings. Then (i) If S is a ﬁeld then ker f is a maximal ideal of R; (ii) If S is an integral domain then ker f is a prime ideal of R. Proof. By the ﬁrst isomorphism theorem (2.3.2), S is isomorphic to R/ker f , and the result now follows from (2.4.3) and (2.4.5). ♣ 2.4.8 Example Let Z[X] be the set of all polynomials f (X) = a0 + a1 X + · · · + an X n , n = 0, 1, . . . in the indeterminate X, with integer coeﬃcients. The ideal generated by X, that is, the collection of all multiples of X, is < X >= {f (X) ∈ Z[X] : a0 = 0}. The ideal generated by 2 is < 2 >= {f (X) ∈ Z[X] : all ai are even integers.} Both < X > and < 2 > are proper ideals, since 2 ∈< / X > and X ∈< / 2 >. In fact we can say much more; consider the ring homomorphisms ϕ : Z[X] → Z and ψ : Z[X] → Z2 given by ϕ(f (X)) = a0 and ψ(f (X)) = a0 , where a0 is a0 reduced modulo 2. We will show that both < X > and < 2 > are prime ideals that are not maximal. First note that < X > is prime by (2.4.7), since it is the kernel of ϕ. Then observe that < X > is not maximal because it is properly contained in < 2, X >, the ideal generated by 2 and X. To verify that < 2 > is prime, note that it is the kernel of the homomorphism from Z[X] onto Z2 [X] that takes f (X) to f (X), where the overbar indicates that the coeﬃcients of f (X) are to be reduced modulo 2. Since Z2 [X] is an integral domain (see the comment at the end of (2.1.3)), < 2 > is a prime ideal. Since < 2 > is properly contained in < 2, X >, < 2 > is not maximal. Finally, < 2, X > is a maximal ideal, since ker ψ = {a0 + Xg(X) : a0 is even and g(X) ∈ Z[X]} =< 2, X > . Thus < 2, X > is the kernel of a homomorphism onto a ﬁeld, and the result follows form (2.4.7). Problems For Section 2.4 1. We know from Problem 1 of Section 2.2 that in the ring of integers, all ideals I are of the form < n > for some n ∈ Z, and since n ∈ I implies −n ∈ I, we may take n to be nonnegative. Let < n > be a nontrivial ideal, so that n is a positive integer greater than 1. Show that < n > is a prime ideal if and only if n is a prime number. 2. Let I be a nontrivial prime ideal in the ring of integers. Show that in fact I must be maximal. 3. Let F [[X]] be the ring of formal power series with coeﬃcients in the ﬁeld F (see (2.1.3),

page 13 of Chapter 2

Example 6). Show that < X > is a maximal ideal. 4. Perhaps the result of Problem 3 is a bit puzzling. Why can’t we argue that just as in (2.4.8), < X > is properly contained in < 2, X >, and therefore < X > is not maximal? 5. Let I be a proper ideal of F [[X]]. Show that I ⊆< X >, so that < X > is the unique maximal ideal of F [[X]]. (A commutative ring with a unique maximal ideal is called a local ring.) 6. Show that every ideal of F [[X]] is principal, and speciﬁcally of the form (X n ) for some n = 0, 1, . . .. 7. Let f : R → S be a ring homomorphism, with R and S commutative. If P is a prime ideal of S, show that the preimage f −1 (P ) is a prime ideal of R. 8. Show that the result of Problem 7 does not hold in general when P is a maximal ideal. 9. Show that a prime ideal P cannot be the intersection of two strictly larger ideals I and J. 2.5 Polynomial Rings In this section, all rings are assumed commutative. To see a good reason for this restriction, consider the evaluation map (also called the substitution map) Ex , where x is a ﬁxed element of the ring R. This map assigns to the polynomial a0 + a1 X + · · · + an X n in R[X] the value a0 + a1 x + · · · + an xn in R. It is tempting to say that “obviously”, Ex is a ring homomorphism, but we must be careful. For example, Ex [(a + bX)(c + dX)] = Ex (ac + (ad + bc)X + bdX 2 ) = ac + (ad + bc)x + bdx2 , but Ex (a + bX)Ex (c + dX) = (a + bx)(c + dx) = ac + adx + bxc + bxdx, and these need not be equal if R is not commutative. If f and g are polynomials in R[X], where R is a ﬁeld, ordinary long division allows us to express f as qg + r, where the degree of r is less than the degree of g. (The degree, abbreviated deg, of a polynomial a0 + a1 X + · · · + an X n (with leading coeﬃcient an = 0) is n; it is convenient to deﬁne the degree of the zero polynomial as −∞. We have a similar result over an arbitrary commutative ring, if g is monic, i.e., the leading coeﬃcient of g is 1. For example (with R = Z), we can divide 2X 3 + 10X 2 + 16X + 10 by X 2 + 3X + 5: 2X 3 + 10X 2 + 16X + 10 = 2X(X 2 + 3X + 5) + 4X 2 + 6X + 10. The remainder 4X 2 +6X +10 does not have degree less than 2, so we divide it by X 2 +3X +5: 4X 2 + 6X + 10 = 4(X 2 + 3X + 5) − 6X − 10. Combining the two calculations, we have 2X 3 + 10X 2 + 16X + 10 = (2X + 4)(X 2 + 3X + 5) + (−6X − 10) which is the desired decomposition. 2.5.1 Division Algorithm If f and g are polynomials in R[X], with g monic, there are unique polynomials q and r in R[X] such that f = qg + r and deg r <deg g. If R is a ﬁeld, g can be any nonzero polynomial. Proof. The above procedure, which works in any ring R, shows that q and r exist.If f = qg + r = q1 g + r1 where r and r1 are of degree less than deg g, then g(q − q1 ) = r1 − r. But if q − q1 = 0, then since g is monic, the degree of the left side is at least deg g, while the degree of the right side is less than deg g, a contradiction. Therefore q = q1 , and consequently r = r1 . ♣ 2.5.2 Remainder Theorem If f ∈ R[X] and a ∈ R, then for some unique polynomial q(X) in R[X] we have f (X) = q(X)(X − a) + f (a);

page 14 of Chapter 2

hence f (a) = 0 if and only if X − a divides f (X). Proof. By the division algorithm, we may write f (X) = q(X)(X − a) + r(X) where the degree of r is less than 1, i.e., r is a constant. Apply the evaluation homomorphism X → a to show that r = f (a). ♣ 2.5.3 Theorem If R is an integral domain, then a nonzero polynomial f in R[X] of degree n has at most n roots in R, counting multiplicity. Proof. If f (a1 ) = 0, then by (2.5.2), possibly applied several times, we have f (X) = q1 (X)(X − a1 )n1 , where q1 (a1 ) = 0 and the degree of q1 is n − n1 . If a2 is another root of f , then 0 = f (a2 ) = q1 (a2 )(a2 −a1 )n1 . But a1 = a2 and R is an integral domain, so q1 (a2 ) must be 0, i.e. a2 is a root of q1 (X). Repeating the argument, we have q1 (X) = q2 (X)(X − a2 )n2 , where q2 (a2 ) = 0 and deg q2 = n − n1 − n2 . After n applications of (2.5.2), the quotient becomes constant, and we have f (X) = c(X − a1 )n1 · · · (X − ak )nk where c ∈ R and n1 + · · · + nk = n. Since R is an integral domain, the only possible roots of f are a1 , . . . , ak . ♣ 2.5.4 Example Let R = Z8 , which is not an integral domain. The polynomial f (X) = X 3 has four roots in R, namely 0,2,4 and 6. Problems For Section 2.5 In the following sequence of problems, we review the Euclidean algorithm. Let a and b be positive integers, with a > b. Divide a by b to obtain a = bq1 + r1 with 0 ≤ r1 < b, then divide b by r1 to get b = r1 q2 + r2 with 0 ≤ r2 < r1 , and continue in this fashion until the process terminates: r 1 = r 2 q 3 + r 3 , 0 ≤ r 3 < r2 , .. . rj−2 = rj−1 qj + rj , 0 ≤ rj < rj−1 , rj−1 = rj qj+1 . 1. Show that the greatest common divisor of a and b is the last remainder rj . 2. If d is the greatest common divisor of a and b, show that there are integers x and y such that ax + by = d. 3. Deﬁne three sequences by ri = ri−2 − qi ri−1 xi = xi−2 − qi xi−1 yi = yi−2 − qi yi−1 for i = −1, 0, 1, . . . with initial conditions r−1 = a, r0 = b, x−1 = 1, x0 = 0, y−1 = 0, y0 = 1. (The qi are determined by dividing ri−2 by ri−1 .) Show that we can generate all steps of the algorithm, and at each stage, ri = axi + byi . 4. Use the procedure of Problem 3 (or any other method) to ﬁnd the greatest common divisor d of a = 123 and b = 54, and ﬁnd integers x and y such that ax + by = d. 5. Use Problem 2 to show that Zp is a ﬁeld if and only if p is prime.

page 15 of Chapter 2

If a(X) and b(X) are polynomials with coeﬃcients in a ﬁeld F , the Euclidean algorithm can be used to ﬁnd their greatest common divisor. The previous discussion can be taken over verbatim, except that instead of writing a = q1 b + r1 with 0 ≤ r1 < b, we write a(X) = q1 (X)b(X) + r1 (X) with deg r1 (X) < deg b(X). The greatest common divisor can be deﬁned as the monic polynomial of highest degree that divides both a(X) and b(X). 6. Let f (X) and g(X) be polynomials in F [X], where F is a ﬁeld. Show that the ideal I generated by f (X) and g(X), i.e., the set of all linear combinations a(X)f (X) + b(X)g(X), with a(X), b(X) ∈ F [X], is the principal ideal J =< d(X) > generated by the greatest common divisor d(X) of f (X) and g(X). 7. (Lagrange Interpolation Formula) Let a0 , a1 , . . . , an be distinct points in the ﬁeld F , and deﬁne

X − aj Pi (X) = , i = 0, 1, . . . , n; ai − aj j=i

then Pi (ai ) = 1 and Pi (aj ) = 0 for j = i. If b0 , b1 , . . . , bn are arbitrary elements of F (not necessarily distinct), use the Pi to ﬁnd a polynomial f (X) of degree n or less such that f (ai ) = bi for all i. 8. In Problem 7, show that f (X) is the unique polynomial of degree n or less such that f (ai ) = bi for all i. 9. Suppose that f is a polynomial in F [X], where F is a ﬁeld. If f (a) = 0 for every a ∈ F , it does not in general follow that f is the zero polynomial. Give an example. 10. Give an example of a ﬁeld F for which it does follow that f = 0. 2.6 Unique Factorization If we are asked to ﬁnd the greatest common divisor of two integers, say 72 and 60, one method is to express each integer as a product of primes; thus 72 = 23 × 32 , 60 = 22 ×3×5. The greatest common divisor is the product of terms of the form pe , where for each prime appearing in the factorization, we use the minimum exponent. Thus gcd(72, 60) = 22 × 31 × 50 = 12. (To ﬁnd the least common multiple, we use the maximum exponent: lcm(72, 60) = 23 × 32 × 51 = 360.) The key idea is that every integer (except 0,1 and -1) can be uniquely represented as a product of primes. It is natural to ask whether there are integral domains other than the integers in which unique factorization is possible. We now begin to study this question; throughout this section, all rings are assumed to be integral domains. 2.6.1 Deﬁnitions Recall from (2.1.2) that a unit in a ring R is an element with a multiplicative inverse. The elements a and b are associates if a = ub for some unit u. Let a be a nonzero nonunit; a is said to be irreducible if it cannot be represented as a product of nonunits. In other words, if a = bc, then either b or c must be a unit. Again let a be a nonzero nonunit; a is said to be prime if whenever a divides a product of terms, it must divide one of the factors. In other words, if a divides bc, then a divides b or a divides c (a divides b means that b = ar for some r ∈ R). It follows from the deﬁnition that if p is any nonzero element of R, then p is prime if and only if is a prime ideal. The units of Z are 1 and −1, and the irreducible and the prime elements coincide. But these properties are not the same in an arbitrary integral domain. 2.6.2 Proposition If a is prime, then a is irreducible, but not conversely. Proof. We use the standard notation r|s to indicate that r divides s. Suppose that a is prime, and that a = bc. Then certainly a|bc, so by deﬁnition of prime, a|b or a|c, say a|b.

page 16 of Chapter 2

If b = ad then b = bcd, so cd = 1 and therefore c is a unit. (Note that b cannot be 0, for if so, a = bc = 0, which is not possible since a is prime.) Similarly, if a|c with c = ad then c = bcd, so bd = 1 and b is a unit. Therefore a is irreducible. √ To√give an example of an irreducible element that is not prime, consider R = Z[ 3] = {a + ib 3 : a, b ∈ Z}; in R, 2 is irreducible but not prime. To see this, ﬁrst suppose that we have a factorization of the form √ √ 2 = (a + ib 3)(c + id 3); take complex conjugates to get √ √ 2 = (a − ib 3)(c − id 3). Now multiply these two equations to obtain 4 = (a2 + 3b2 )(c2 + 3d2 ). Each factor on the right must be a divisor of 4, and there is no way that a2 + 3b2 can be 2. Thus one of the factors must be 4 and the other must be 1. If, say, a2 + 3b2 = 1, then a = ±1 and b = 0. Thus in the original factorization of 2, one √ of the√factors must be a unit, so 2 is irreducible. Finally, 2 divides the product (1 + i 3)(1 − i 3) (= 4), so if 2 were prime, it would divide one of the factors, which means that 2 divides 1, a contradiction since 1/2 is not an integer. ♣ The distinction between irreducible and prime elements disappears in the presence of unique factorization. 2.6.3 Deﬁnition A unique factorization domain (UFD) is an integral domain R satisfying the following properties: (UF1) Every nonzero element a in R can be expressed as a = up1 · · · pn , where u is a unit and the pi are irreducible. (UF2): If a has another factorization, say a = vq1 · · · qm , where v is a unit and the qi are irreducible, then n = m and, after reordering if necessary, pi and qi are associates for each i. Property UF1 asserts the existence of a factorization into irreducibles, and U F 2 asserts uniqueness. 2.6.4 Proposition In a unique factorization domain, a is irreducible if and only if a is prime. Proof. By (2.6.2), prime implies irreducible, so assume a irreducible, and let a divide bc. Then we have ad = bc for some d ∈ R. We factor d, b and c into irreducibles to obtain aud1 · · · dr = vb1 · · · bs wc1 · · · ct where u, v and w are units and the di , bi and ci are irreducible. By uniqueness of factorization, a, which is irreducible, must be an associate of some bi or ci . Thus a divides b or a divides c. ♣ 2.6.5 Deﬁnitions and Comments Let A be a nonempty subset of R, with 0 ∈ / A. The element d is a greatest common divisor (gcd) of A if d divides each a in A, and whenever e divides each a in A, we have e|d. If d is another gcd of A, we have d|d and d |d, so that d and d are associates. We will allow ourselves to speak of “the” greatest common divisor, suppressing but not forgetting that the gcd is determined up to multiplication by a unit.

page 17 of Chapter 2

The elements of A are said to be relatively prime (or the set A is said to be relatively prime) if 1 is a greatest common divisor of A. The nonzero element m is a least common multiple (lcm) of A if each a in A divides m, and whenever a|e for each a in A, we have m|e. Greatest common divisors and least common multiples always exist for ﬁnite subsets of a UFD; they may be found by the technique discussed at the beginning of this section. We will often use the fact that for any a, b ∈ R, we have a|b if and only if ⊆< a >. This follows because a|b means that b = ac for some c ∈ R. For short, divides means contains. It would be useful to be able to recognize when an integral domain is a UFD. The following criterion is quite abstract, but it will help us to generate some explicit examples. 2.6.6 Theorem Let R be an integral domain. Then: (1) If R is a UFD then R satisﬁes the ascending chain condition (acc) on principal ideals, in other words, if a1 , a2 , . . . belong to R and < a1 >⊆< a2 >⊆ . . ., then the sequence eventually stabilizes, that is, for some n we have < an >=< an+1 >=< an+2 >= . . .. (2) If R satisﬁes the ascending chain condition on principal ideals, then R satisﬁes UF1, that is, every nonzero element of R can be factored into irreducibles. (3) If R satisﬁes UF1 and in addition, every irreducible element of R is prime, then R is a UFD. Thus R is a UFD if and only if R satisﬁes the ascending chain condition on principal ideals and every irreducible element of R is prime. Proof. (1) If < a1 >⊆< a2 >⊆ . . . then ai+1 |ai for all i. Therefore the prime factors of ai+1 consist of some (or all) of the prime factors of ai . Multiplicity is taken into account here; for example, if p3 is a factor of ai , then pk will be a factor of ai+1 for some k ∈ {0, 1, 2, 3}. Since a1 has only ﬁnitely many prime factors, there will come a time when the prime factors are the same from that point on, that is, < an >=< an+1 >= . . .. (2) Let a1 be any nonzero element. If a1 is irreducible, we are ﬁnished, so let a1 = a2 b2 where neither a2 nor b2 is a unit. If both a2 and b2 are irreducible, we are ﬁnished, so we can assume that one of them, say a2 , is not irreducible. Since a2 divides a1 we have < a1 >⊆< a2 >, and in fact the inclusion is proper because a2 ∈< / a1 >. (If a2 = ca1 then a1 = a2 b2 = ca1 b2 , so b2 is a unit, a contradiction.) Continuing, we have a2 = a3 b3 where neither a3 nor b3 is a unit, and if, say, a3 is not irreducible, we ﬁnd that < a2 >⊂< a3 >. If a1 cannot be factored into irreducibles, we obtain, by an inductive argument, a strictly increasing chain < a1 >⊂< a2 >⊂ . . . of principal ideals. (3) Suppose that a = up1 p2 · · · pn = vq1 q2 · · · qm where the pi and qi are irreducible and u and v are units. Then p1 is a prime divisor of vq1 · · · qm , so p1 divides one of the qi , say q1 . But q1 is irreducible, and therefore p1 and q1 are associates. Thus we have, up to multiplication by units, p2 . . . pn = q2 · · · qm . By an inductive argument, we must have m = n, and after reordering, pi and qi are associates for each i. ♣ We now give a basic suﬃcient condition for an integral domain to be a UFD. 2.6.7 Deﬁnition A principal ideal domain (PID) is an integral domain in which every ideal is principal, that is, generated by a single element. 2.6.8 Theorem Every principal ideal domain is a unique factorization domain. For short, PID implies UFD. Proof. If< a1 >⊆< a2 >⊆ . . ., let I = ∪i < ai >. Then I is an ideal, necessarily principal by hypothesis. If I = then b belongs to some < an >, so I ⊆< an >. Thus if i ≥ n we have < ai >⊆ I ⊆< an >⊆< ai >. Therefore < ai >=< an > for all i ≥ n, so that R satisﬁes the acc on principal ideals.

page 18 of Chapter 2

Now suppose that a is irreducible. Then < a > is a proper ideal, for if < a >= R then 1 ∈< a >, so that a is a unit. By the acc on principal ideals, < a > is contained in a maximal ideal I. (Note that we need not appeal to the general result (2.4.2), which uses Zorn’s lemma.) If I = then b divides the irreducible element a, and b is not a unit since I is proper. Thus a and b are associates, so < a >== I. But I, a maximal ideal, is prime by (2.4.6), hence a is prime. The result follows from (2.6.6). ♣ The following result gives a criterion for a UFD to be a PID. (Terminology: the zero ideal is {0}; a nonzero ideal is one that is not {0}.) 2.6.9 Theorem R is a PID iﬀ R is a UFD and every nonzero prime ideal of R is maximal. Proof. Assume R is a PID; then R is a UFD by (2.6.8). If is a nonzero prime ideal of R, then is contained in the maximal ideal < q >, so that q divides the prime p. Since a maximal ideal must be proper, q cannot be a unit, so that p and q are associates. But then =< q > and is maximal. The proof of the converse is given in the exercises. ♣ Problems For Section 2.6 The problems in this section form a project designed to prove that if R is a UFD and every nonzero prime ideal of R is maximal, then R is a PID. 1. Let I be an ideal of R; since {0} is principal, we can assume that I = {0}. Since R is a UFD, every nonzero element of I can be written as up1 · · · pt where u is a unit and the pi are irreducible, hence prime. Let r = r(I) be the minimum such t. We are going to prove by induction on r that I is principal. If r = 0, show that I =< 1 >= R. 2. If the result holds for all r < n, let r = n, with up1 · · · pn ∈ I, hence p1 · · · pn ∈ I. Since p1 is prime, < p1 > is a prime ideal, necessarily maximal by hypothesis. By (2.4.3), R/ is a ﬁeld. If b belongs to I but not to < p1 >, show that for some c ∈ R we have bc − 1 ∈< p1 >. 3. By Problem 2, bc − dp1 = 1 for some d ∈ R. Show that this implies that p2 · · · pn ∈ I, which contradicts the minimality of n. Thus if b belongs to I, it must also belong to < p1 >, that is, I ⊆< p1 >. 4. Deﬁne J = {x ∈ R : xp1 ∈ I}, and show that J is an ideal. 5. Show that Jp1 = I. 6. Since p1 · · · pn = (p2 · · · pn )p1 ∈ I, we have p2 · · · pn ∈ J. Use the induction hypothesis to conclude that I is principal. 7. Let p and q be prime elements in the integral domain R, and let P = and Q =< q > be the corresponding prime ideals. Show that it is not possible for P to be a proper subset of Q. 8. If R is a UFD and P is a nonzero prime ideal of R, show that P contains a nonzero principal prime ideal. 2.7 Principal Ideal Domains and Euclidean Domains In Section 2.6, we found that a principal ideal domain is a unique factorization domain, and this exhibits a class of rings in which unique factorization occurs. We now study some properties of PID’s, and show that any integral domain in which the Euclidean algorithm works is a PID. If I is an ideal in Z, in fact if I is simply an additive subgroup of Z, then I consists of all multiples of some positive integer n; see Section 1.1, Problem 6. Thus Z is a PID. Now suppose that A is a nonempty subset of the PID R. The ideal < A > generated by A consists of all ﬁnite sums ri ai with ri ∈ R and ai ∈ A; see (2.2.7). We show that if d is a greatest common divisor of A, then d generates A, and conversely.

page 19 of Chapter 2

2.7.1 Proposition Let R be a PID, with A a nonempty subset of R. Then d is a greatest common divisor of A if and only if d is a generator of < A >. Proof. Let d be a gcd of A, and assume that < A >=. Then d divides every a ∈ A, so d divides all ﬁnite sums ri ai , in particular d divides b, hence ⊆< d >, that is, < A >⊆< d >. But if a ∈ A then a ∈, so that b divides a. Since d is a gcd of A, it follows that b divides d, so < d > is contained in =< A >. We conclude that < A >=< d >, proving that d is a generator of < A >. Conversely, assume that d generates < A>. If a ∈ A then a is a multiple of d, so that d|a. Since (by (2.2.7)) d can be expressed as ri ai , any element that divides everything in A divides d, so that d is a gcd of A. ♣ 2.7.2 Corollary If d is a gcd of A, where A isa nonempty subset of the PID R, then d can be expressed as a ﬁnite linear combination ri ai of elements of A with coeﬃcients in R. Proof. By (2.7.1), d ∈< A >, and the result follows from (2.2.7). ♣ As a special case, we have the familiar result that the greatest common divisor of two integers a and b can be expressed as ax + by for some integers x and y. The Euclidean algorithm in Z is based on the division algorithm: if a and b are integers and b = 0, then a can be divided by b to produce a quotient and remainder. Speciﬁcally, we have a = bq + r for some q, r ∈ Z with |r| < |b|. The Euclidean algorithm performs equally well for polynomials with coeﬃcients in a ﬁeld; the absolute value of an integer is replaced by the degree of a polynomial. It is possible to isolate the key property that makes the Euclidean algorithm work. 2.7.3 Deﬁnition Let R be an integral domain. R is said to be a Euclidean domain (ED) if there is a function Ψ from R \ {0} to the nonnegative integers satisfying the following property: If a and b are elements of R, with b = 0, then a can be expressed as bq + r where either r = 0 or Ψ(r) < Ψ(b). We can replace“r = 0 or Ψ(r) < Ψ(b)” by simply “Ψ(r) < Ψ(b)” if we deﬁne Ψ(0) to be −∞. In any Euclidean domain, we may use the Euclidean algorithm to ﬁnd the greatest common divisor of two elements; see the Problems in Section 2.5 for a discussion of the procedure in Z and in F [X], where F is a ﬁeld. A Euclidean domain is automatically a principal ideal domain, as we now prove. 2.7.4 Theorem If R is a Euclidean domain, then R is a principal ideal domain. For short, ED implies PID. Proof. Let I be an ideal of R. If I = {0} then I is principal, so assume I = {0}. Then {Ψ(b) : b ∈ I, b = 0} is a nonempty set of nonnegative integers, and therefore has a smallest element n. Let b be any element of I such that Ψ(b) = n; we claim that I =. For if a belongs to I then we have a = bq + r where r = 0 or Ψ(r) < Ψ(b). Now r = a − bq ∈ I (because a and b belong to I), so if r = 0 then Ψ(r) < Ψ(b) is impossible by minimality of Ψ(b). Thus b is a generator of I. ♣ The most familiar Euclidean domains are Z and F [X], with F a ﬁeld. We now examine some less familiar cases. √ √ 2.7.5 Example Let Z[ d] be√the ring of all elements a + b d, where a, b ∈ Z. If d = −2, −1, 2 or 3, we claim that Z[ d] is a Euclidean domain with √ Ψ(a + b d) = |a2 − db2 |.

page 20 of Chapter 2

√ √ Since the a + b d are real or complex numbers, there are no zero divisors, √ √ and Z[ d] is an integral domain. Let α, β ∈ Z[ d], β = 0, and divide α by β to get x + y d. Unfortunately, x and y need not be integers, but at least they are rational numbers. We can ﬁnd integers reasonably close to x and y; let x0 and y0 be integers such that |x − x0 | and |y − y0 | are at most 1/2. Let √ √ √ q = x0 + y0 d, r = β((x − x0 ) + (y − y0 ) d); then βq + r = β(x + y d) = α. We must show that Ψ(r) < Ψ(β). Now √ √ √ Ψ(a + b d) = |(a + b d)(a − b d)|, √ and it follows (Problem 4) that for all γ, δ ∈ Z[ d] we have Ψ(γδ) = Ψ(γ)Ψ(δ). (When d = −1, this says that the magnitude of the product of two complex numbers is the product of the magnitudes.) Thus Ψ(r) = Ψ(β)[(x − x0 )2 − d(y − y0 )2 ], and the factor in brackets is at most 14 + |d|( 14 ) ≤ 14 + 34 = 1. The only possibility for equality occurs when d = 3 (d = −3 is excluded by hypothesis) and |x − x0 | = |y − y0 | = 12 . But in this case, the √ factor in brackets is | 14 − 3( 14 )| = 12 < 1. We have shown that Ψ(r) < Ψ(β), so that Z[ d] is a Euclidean domain. √ When d = −1, we obtain the Gaussian integers a + bi, a, b ∈ Z, i = −1. Problems For Section 2.7 1. Let A = {a1 , . . . , an } be a ﬁnite subset of the PID R. Show that m is a least common multiple of A iﬀ m is a generator of the ideal ∩ni=1 < ai >. 2. Find the gcd of 11 + 3i and 8 − i in the ring of Gaussian integers. 3. Suppose that R is a Euclidean domain in which Ψ(a) ≤ Ψ(ab) for all nonzero elements a, b ∈ R. Show that a unit in R. √ Ψ(a) ≥ Ψ(1), with equality if and only if a is √ 4. Let R = Z[ d], where d is any integer, and deﬁne Ψ(a + b d) = |a2 − db2 |. Show that for all nonzero α and β, Ψ(αβ) = Ψ(α)Ψ(β), and if d is not a perfect square, then Ψ(α) ≤ Ψ(αβ).√ 5. Let R = Z[ d] where d is not a perfect square. Show that 2 is not prime in R. (Show that 2 divides d2 − d.) √ 6. If d is a negative integer with d ≤ −3, show that 2 is irreducible in Z[ d]. √ 7. Let R = Z[ d] where d is a negative integer. We know (see (2.7.5)) that R is an ED, hence a PID and a UFD, for d = −1 and d = −2. Show that for d ≤ −3, R is not a UFD. 8. Find the least common multiple of 11 + 3i and 8 − i in the ring of Gaussian integers. 9. If α = a + bi is a Gaussian integer, let ψ(α) = a2 + b2 as in (2.7.5). If Ψ(α) is prime in Z, show that α is prime in Z[i]. 2.8 Rings of Fractions It was recognized quite early in mathematical history that the integers have a significant defect: the quotient of two integers need not be an integer. In such a situation a mathematician is likely to say“I want to be able to divide one integer by another, and I will”. This will be legal if the computation takes place in a ﬁeld F containing the integers Z. Any such ﬁeld will do, since if a and b belong to F and b = 0, then a/b ∈ F . How do we know that a suitable F exists? With hindsight we can take F to be the rationals Q, the reals R, or the complex numbers C. In fact, Q is the smallest ﬁeld containing Z, since any ﬁeld F ⊇ Z contains a/b for all a, b ∈ Z, b = 0, and consequently F ⊇ Q. The same process that leads from the integers to the rationals allows us to construct, for an arbitrary integral domain D, a ﬁeld F whose elements are (essentially) fractions

page 21 of Chapter 2

a/b, a, b ∈ D, b = 0. F is called the ﬁeld of fractions or quotient ﬁeld of D. The mathematician’s instinct to generalize then leads to the following question: If R is an arbitrary commutative ring (not necessarily an integral domain), can we still form fractions with numerator and denominator in R? Diﬃculties quickly arise; for example, how do we make sense of ab dc when bd = 0? Some restriction must be placed on the allowable denominators, and we will describe a successful approach shortly. Our present interest in is the ﬁeld of fractions of an integral domain, but later we will need the more general development. Since the ideas are very similar, we will give the general construction now. 2.8.1 Deﬁnitions and Comments Let S be a subset of the ring R; we say that S is multiplicative if 0 ∈ / S, 1 ∈ S, and whenever a and b belong to S, we have ab ∈ S. We can merge the last two requirements by stating that S is closed under multiplication, if we regard 1 as the empty product. Here are some standard examples. (1) S = the set of all nonzero elements of an integral domain. (2) S = the set of all elements of a commutative ring R that are not zero divisors. (3) S = R \ P , where P is a prime ideal of the commutative ring R. If S is a multiplicative subset of the commutative ring R, we deﬁne the following equivalence relation on R × S: (a, b) ∼ (c, d) iﬀ for some s ∈ S we have s(ad − bc) = 0. If we are constructing the ﬁeld of fractions of an integral domain, then (a, b) is our ﬁrst approximation to a/b. Also, since the elements s ∈ S are never 0 and R has no zero divisors, we have (a, b) ∼ (c, d) iﬀ ad = bc, and this should certainly be equivalent to a/b = c/d. Let us check that we have a legal equivalence relation. [Commutativity of multiplication will be used many times to slide an element to a more desirable location in a formula. There is a theory of rings of fractions in the noncommutative case, but we will not need the results, and in view of the serious technical diﬃculties that arise, we will not discuss this area.] Reﬂexivity and symmetry follow directly from the deﬁnition. For transitivity, suppose that (a, b) ∼ (c, d) and (c, d) ∼ (e, f ). Then for some elements s and t in S we have s(ad − bc) = 0 and t(cf − de) = 0. Multiply the ﬁrst equation by tf and the second by sb, and add the results to get std(af − be) = 0, which implies that (a, b) ∼ (e, f ), proving transitivity. If a ∈ R and b ∈ S, we deﬁne the fraction ab to be the equivalence class of the pair (a, b). The set of all equivalence classes is denoted by S −1 R, and is called (in view of what we are about to prove) the ring of fractions of R by S. The term localization of R by S is also used, because it will turn out that in Examples (1) and (3) above, S −1 R is a local ring (see Section 2.4, Problem 5). We now make the set of fractions into a ring in a natural way.

page 22 of Chapter 2 ad+bc bd multiplication: ab dc = ac bd 0 additive identity: 1 (= 0s for any s ∈ S) additive inverse: −( ab ) = −a b 1 multiplicative identity: 1 (= ss for any s

addition:

a b

+

c d

=

∈ S)

2.8.2 Theorem With the above deﬁnitions, S −1 R is a commutative ring. If R is an integral domain, so is S −1 R. If R is an integral domain and S = R \ {0}, then S −1 R is a ﬁeld (the ﬁeld of fractions or quotient ﬁeld of R). Proof. First we show that addition is well-deﬁned. If a1 /b1 = c1 /d1 and a2 /b2 = c2 /d2 , then for some s, t ∈ S we have s(a1 d1 − b1 c1 ) = 0 and t(a2 d2 − b2 c2 ) = 0

(1)

Multiply the ﬁrst equation of (1) by tb2 d2 and the second equation by sb1 d1 , and add the results to get st[(a1 b2 + a2 b1 )d1 d2 − (c1 d2 + c2 d1 )b1 b2 ] = 0. Thus

in other words,

a1 b2 + a2 b1 c1 d2 + c2 d1 = , b1 b2 d1 d2 a1 a2 c1 c2 + = + b1 b2 d1 d2

so that addition of fractions does not depend on the particular representative of an equivalence class. Now we show that multiplication is well-deﬁned. We follow the above computation as far as (1), but now we multiply the ﬁrst equation by ta2 d2 , the second by sc1 b1 , and add. The result is st[a1 a2 d1 d2 − b1 b2 c1 c2 ] = 0 which implies that

a1 a2 c1 c2 = , b1 b2 d1 d2

as desired. We now know that the fractions in S −1 R can be added and multiplied in exactly the same way as ratios of integers, so checking the deﬁning properties of a commutative ring essentially amounts to checking that the rational numbers form a commutative ring; see Problems 3 and 4 for some examples. Now assume that R is an integral domain. It follows that if a/b is zero in S −1 R, i.e., a/b = 0/1, then a = 0 in R. (For some s ∈ S we have sa = 0, and since R is an integral domain and s = 0, we must have a = 0.) Thus if ab dc = 0, then ac = 0, so either a or c is 0, and consequently either a/b or c/d is zero. Therefore S −1 R is an integral domain. If R is an integral domain and S = R \ {0}, let a/b be a nonzero element of S −1 R. Then both a and b are nonzero, so that a, b ∈ S. By deﬁnition of multiplication we have a b 1 −1 R is a ﬁeld. ♣ b a = 1 . Thus a/b has a multiplicative inverse, so that S When we go from the integers to the rational numbers, we don’t lose the integers in the process, in other words, the rationals contain a copy of the integers, namely, the rationals of the form a/1, a ∈ Z. So a natural question is whether S −1 R contains a copy of R. 2.8.3 Proposition Deﬁne f : R → S −1 R by f (a) = a/1. Then f is a ring homomorphism. If S has no zero divisors then f is a monomorphism, and we say that R can be embedded in S −1 R. In particular,

page 23 of Chapter 2

(i) A commutative ring R can be embedded in its complete (or full) ring of fractions (S −1 R, where S consists of all non-divisors of zero in R). (ii) An integral domain can be embedded in its quotient ﬁeld. ab Proof. We have f (a + b) = a+b = a1 + 1b = f (a) + f (b), f (ab) = ab 1 1 = 1 1 = f (a)f (b), 1 and f (1) = 1 , proving that f is a ring homomorphism. If S has no zero divisors and f (a) = a/1 = 0/1, then for some s ∈ S we have sa = 0, and since s cannot be a zero divisor, we have a = 0. Thus f is a monomorphism. ♣

2.8.4 Corollary The quotient ﬁeld F of an integral domain R is the smallest ﬁeld containing R. Proof. By (2.8.3), we may regard R as a subset of F , so that F is a ﬁeld containing R. But if L is any ﬁeld containing R, then all fractions a/b, a, b ∈ R, must belong to L. Thus F ⊆ L. ♣ Problems For Section 2.8 1. If the integral domain D is in fact a ﬁeld, what is the quotient ﬁeld of D? 2. If D is the set F [X] of all polynomials over a ﬁeld F , what is the quotient ﬁeld of D? 3. Give a detailed proof that addition in a ring of fractions is associative. 4. Give a detailed proof that the distributive laws hold in a ring of fractions. 5. Let R be an integral domain with quotient ﬁeld F , and let h be a ring monomorphism from R to a ﬁeld L. Show that h has a unique extension to a monomorphism from F to L. 6. Let h be the ring homomorphism from Z to Zp , p prime, given by h(x) = x mod p. Why can’t the analysis of Problem 5 be used to show that h extends to a monomorphism of the rationals to Zp ? (This can’t possibly work since Zp is ﬁnite, but what goes wrong?) 7. Let S be a multiplicative subset of the commutative ring R, with f : R → S −1 R deﬁned by f (a) = a/1. If g is a ring homomorphism from R to a commutative ring R and g(s) is a unit in R for each s ∈ S, we wish to ﬁnd a ring homomorphism g : S −1 R → R such that g(f (a)) = g(a) for every a ∈ R. Thus the diagram below is commutative.

f

R

g →

R

↓

g

S −1 R Show that there is only one conceivable way to deﬁne g. 8. Show that the mapping you have deﬁned in Problem 7 is a well-deﬁned ring homomorphism. 2.9 Irreducible Polynomials 2.9.1 Deﬁnitions and Comments In (2.6.1) we deﬁned an irreducible element of a ring; it is a nonzero nonunit which cannot be represented as a product of nonunits. If R is an integral domain, we will refer to an irreducible element of R[X] as an irreducible polynomial. Now in F [X], where F is a ﬁeld, the units are simply the nonzero elements of F (Section 2.1, Problem 2). Thus in this case, an irreducible element is a polynomial of degree at least 1 that cannot be factored into two polynomials of lower degree. A polynomial that is not irreducible is said to be reducible or factorable. For example, X 2 + 1, regarded as an element of R[X], where R is the ﬁeld of real numbers, is irreducible, but if we replace √ R by the larger ﬁeld C of complex numbers, X 2 + 1 is factorable as (X − i)(X + i), i = −1. We say that X 2 + 1 is irreducible over R but not irreducible over C. Now consider D[X], where D is a unique factorization domain but not necessarily a ﬁeld, for example, D = Z. The polynomial 12X +18 is not an irreducible element of Z[X] because it can be factored as the product of the two nonunits 6 and 2X + 3. It is convenient to

page 24 of Chapter 2

factor out the greatest common divisor of the coeﬃcients (6 in this case). The result is a primitive polynomial, one whose content (gcd of coeﬃcients) is 1. A primitive polynomial will be irreducible if and only if it cannot be factored into two polynomials of lower degree. In this section, we will compare irreducibility over a unique factorization domain D and irreducibility over the quotient ﬁeld F of D. Here is the key result. 2.9.2 Proposition Let D be a unique factorization domain with quotient ﬁeld F . Suppose that f is a polynomial in D[X] and that f can be factored as gh, where g and h belong to F [X]. Then there is a nonzero element λ ∈ F such that λg ∈ D[X] and λ−1 h ∈ D[X]. Thus if f is factorable over F , then it is factorable over D. Equivalently, if f is irreducible over D, then f is irreducible over F . Proof. The coeﬃcients of g and h are quotients of elements of D. If a is the least common denominator for g (technically, the least common multiple of the denominators of the coefﬁcients of g), then g ∗ = ag ∈ D[X]. Similarly we have h∗ = bh ∈ D[X]. Thus abf = g ∗ h∗ with g ∗ , h∗ ∈ D[X] and c = ab ∈ D. Now if p is a prime factor of c, we will show that either p divides all coeﬃcients of g ∗ or p divides all coeﬃcients of h∗ . We do this for all prime factors of c to get f = g0 h0 with g0 , h0 ∈ D[X]. Since going from g to g0 involves only multiplication or division by nonzero constants in D, we have g0 = λg for some nonzero λ ∈ F . But then h0 = λ−1 h, as desired. Thus let g ∗ (X) = g0 + g1 X + · · · + gs X s , h∗ (X) = h0 + h1 X + · · · + ht X t . Since p is a prime factor of c = ab and abf = g ∗ h∗ , p must divide all coeﬃcients of g ∗ h∗ . If p does not divide every gi and p does not divide every hi , let gu and hv be the coeﬃcients of minimum index not divisible by p. Then the coeﬃcient of X u+v in g ∗ h∗ is g0 hu+v + g1 hu+v−1 + · · · + gu hv + · · · + gu+v−1 h1 + gu+v h0 . But by choice of u and v, p divides every term of this expression except gu hv , so that p cannot divide the entire expression. So there is a coeﬃcient of g ∗ h∗ not divisible by p, a contradiction. ♣ The technique of the above proof yields the following result. 2.9.3 Gauss’ Lemma Let f and g be polynomials in D[X], where D is a unique factorization domain. If c denotes content, then c(f g) = c(f )c(g). In particular, the product of two primitive polynomials is primitive. Proof. By deﬁnition of content we may write f = c(f )f ∗ and g = c(g)g ∗ where f ∗ and g ∗ are primitive. Thus f g = c(f )c(g)f ∗ g ∗ . It follows that c(f )c(g) divides every coeﬃcient of f g, so c(f )c(g) divides c(f g). Now let p be any prime factor of c(f g); then p divides c(f )c(g)f ∗ g ∗ , and the proof of (2.9.2) shows that either p divides every coeﬃcient of c(f )f ∗ or p divides every coeﬃcient of c(g)g ∗ . If, say, p divides every coeﬃcient of c(f )f ∗ , then (since p is prime) either p divides c(f ) or p divides every coeﬃcient of f ∗ . But f ∗ is primitive, so that p divides c(f ), hence p divides c(f )c(g). We conclude that c(f g) divides c(f )c(g), and the result follows. ♣ 2.9.4 Corollary (of the proof of (2.9.3)) If h ∈ D[X] and h = ah∗ where h∗ is primitive, then a must be the content of h. Proof. Since a divides every coeﬃcient of h, a must divide c(h). If p is any prime factor of c(h), then p divides every coeﬃcient of ah∗ , and as in (2.9.3), either p divides a or p divides every coeﬃcient of h∗ , which is impossible by primitivity of h∗ . Thus c(h) divides a, and the result follows. ♣

page 25 of Chapter 2

Proposition 2.9.2 yields a precise statement comparing irreducibility over D with irreducibility over F . 2.9.5 Proposition Let D be a unique factorization domain with quotient ﬁeld F . If f ∈ D[X], then f is irreducible over D if and only if f is primitive and irreducible over F . Proof. If f is irreducible over D, then f is irreducible over F by (2.9.2). If f is not primitive, then f = c(f )f ∗ where f ∗ is primitive and c(f ) is not a unit. This contradicts the irreducibility of f over D. Conversely, if f = gh is a factorization of the primitive polynomial f over D, then g and h must be of degree at least 1. Thus neither g nor h is a unit in F [X], so f = gh is a factorization of f over F . ♣ Here is another basic application of (2.9.2). 2.9.6 Theorem If R is a unique factorization domain, so is R[X]. Proof. If f ∈ R[X], f = 0, then f can be factored over the quotient ﬁeld F as f = f1 f2 · · · fk , where the fi are irreducible polynomials in F [X]. (Recall that F [X] is a Euclidean domain, hence a unique factorization domain.) By (2.9.2), for some nonzero λ1 ∈ F we may write −1 f = (λ1 f1 )(λ−1 1 f2 · · · fk ) with λ1 f1 and λ1 f2 · · · fk in R[X]. Again by (2.9.2), we have −1 −1 −1 −1 −1 λ1 f2 · · · fk = f2 λ1 f3 · · · fk = (λ2 f2 )(λ2 λ1 f3 · · · fk ) with λ2 f2 and λ−1 2 λ1 f3 · · · fk ∈ k R[X]. Continuing inductively, we express f as i=1 λi fi where the λi fi are in R[X] and are irreducible over F . But λi fi is the product of its content and a primitive polynomial (which is irreducible over F , hence over R by (2.9.5)). Furthermore, the content is either a unit or a product of irreducible elements of the UFD R, and these elements are irreducible in R[X] as well. This establishes the existence of a factorization into irreducibles. Now suppose that f = g1 · · · gr = h1 · · · hs , where the gi and hi are irreducible polynomials in R[X]. By (2.9.5), the gi and hi are irreducible over F , and since F [X] is a UFD, we have r = s and, after reordering if necessary, gi and hi are associates (in F [X]) for each i. Now gi = ci hi for some constant ci ∈ F , and we have ci = ai /bi with ai , bi ∈ R. Thus bi gi = ai hi , with gi and hi primitive by (2.9.5). By (2.9.4), bi gi has content bi and ai hi has content ai . Therefore ai and bi are associates, which makes ci a unit in R, which in turn makes gi and hi associates in R[X], proving uniqueness of factorization. ♣ The following result is often used to establish irreducibility of a polynomial. 2.9.7 Eisenstein’s Irreducibility Criterion Let R be a UFD with quotient ﬁeld F , and let f (X) = an X n + · · · + a1 X + a0 be a polynomial in R[X], with n ≥ 1 and an = 0. If p is prime in R, p divides ai for 0 ≤ i < n, but p does not divide an and p2 does not divide a0 , then f is irreducible over F . Thus by (2.9.5), if f is primitive then f is irreducible over R. Proof. If we divide f by its content to produce a primitive polynomial f ∗ , the hypothesis still holds for f ∗ . (Since p does not divide an , it is not a prime factor of c(f ), so it must divide the ith coeﬃcient of f ∗ for 0 ≤ i < n.) If we can prove that f ∗ is irreducible over R, then by (2.9.5), f ∗ is irreducible over F , and therefore so is f . Thus we may assume without loss of generality that f is primitive, and prove that f is irreducible over R. Assume that f = gh, with g(X) = g0 + · · · + gr X r and h(X) = h0 + · · · + hs X s . If r = 0 then g0 divides all coeﬃcients ai of f , so g0 divides c(f ), hence g(= g0 ) is a unit. Thus we may assume that r ≥ 1, and similarly s ≥ 1. By hypothesis, p divides a0 = g0 h0 but p2 does not divide a0 , so p cannot divide both g0 and h0 . Assume that p fails to divide h0 , so that p divides g0 ; the argument is symmetrical in the other case. Now gr hs = an , and by hypothesis, p does not divide an , so that p does not divide gr . Let i be the smallest integer such that p does not divide gi ; then 1 ≤ i ≤ r < n (since r + s = n and s ≥ 1). Now ai = g0 hi + g1 hi−1 + · · · + gi h0 and by choice of i, p divides g0 , . . . , gi−1 . But p divides the entire sum ai , so p must divide the last term gi h0 . Consequently, either p divides gi , which contradicts the choice of i, or p

page 26 of Chapter 2

divides h0 , which contradicts our earlier assumption. Thus there can be no factorization of f as a product of polynomials of lower degree, in other words, f is irreducible over R. ♣ Problems For Section 2.9 1. (The rational root test, which can be useful in factoring a polynomial over Q.) Let f (X) = an X n + · · · + a1 X + a0 ∈ Z[X]. If f has a rational root u/v where u and v are relatively prime integers and v = 0, show that v divides an and u divides a0 . 2. Show that for every positive integer n, there is at least one irreducible polynomial of degree n over the integers. 3. If f (X) ∈ Z[X] and p is prime, we can reduce all coeﬃcients of f modulo p to obtain a new polynomial fp (X) ∈ Zp [X]. If f is factorable over Z, then fp is factorable over Zp . Therefore if fp is irreducible over Zp , then f is irreducible over Z. Use this idea to show that the polynomial X 3 + 27X 2 + 5X + 97 is irreducible over Z. (Note that Eisenstein does not apply.) 4. If we make a change of variable X = Y + c in the polynomial f (X), the result is a new polynomial g(Y ) = f (Y + c). If g is factorable over Z, so is f since f (X) = g(X − c). Thus if f is irreducible over Z, so is g. Use this idea to show that X 4 + 4X 3 + 6X 2 + 4X + 4 is irreducible over Z. 5. Show that in Z[X], the ideal < n, X >, n ≥ 2, is not principal, and therefore Z[X] is a UFD that is not a PID. 6. Show that if F is a ﬁeld, then F [X, Y ], the set of all polynomials aij X i Y j , aij ∈ F , is not a PID since the ideal < X, Y > is not principal. 7. Let f (X, Y ) = X 2 + Y 2 + 1 ∈ C[X, Y ], where C is the ﬁeld of complex numbers. Write f as Y 2 + (X 2 + 1) and use Eisenstein’s criterion to show that f is irreducible over C. 8. Show that f (X, Y ) = X 3 + Y 3 + 1 is irreducible over C.

page 1 of Chapter 3

CHAPTER 3 FIELD FUNDAMENTALS 3.1 Field Extensions If F is a ﬁeld and F [X] is the set of all polynomials over F, that is, polynomials with coeﬃcients in F , we know that F [X] is a Euclidean domain, and therefore a principal ideal domain and a unique factorization domain (see Sections 2.6 and 2.7). Thus any nonzero polynomial f in F [X] can be factored uniquely as a product of irreducible polynomials. Any root of f must be a root of one of the irreducible factors, but at this point we have no concrete information about the existence of roots and how they might be found. For example, X 2 + 1 has no real roots, but if we consider the larger ﬁeld of complex numbers, we get two roots, +i and −i. It appears that the process of passing to a larger ﬁeld may help produce roots, and this turns out to be correct. 3.1.1 Deﬁnitions If F and E are ﬁelds and F ⊆ E, we say that E is an extension of F , and we write F ≤ E, or sometimes E/F . If E is an extension of F , then in particular E is an abelian group under addition, and we may multiply the “vector” x ∈ E by the “scalar” λ ∈ F , and the axioms of a vector space are satisﬁed. Thus if F ≤ E, then E is a vector space over F . The dimension of this vector space is called the degree of the extension, written [E : F ]. If [E : F ] = n < ∞, we say that E is a ﬁnite extension of F , or that the extension E/F is ﬁnite, or that E is of degree n over F . If f is a nonconstant polynomial over the ﬁeld F , and f has no roots in F , we can always produce a root of f in an extension ﬁeld of F . We do this after a preliminary result. 3.1.2 Lemma Let f : F → E be a homomorphism of ﬁelds, i.e., f (a + b) = f (a) + f (b), f (ab) = f (a)f (b) (all a, b ∈ F ), and f (1F ) = 1E . Then f is a monomorphism. Proof. First note that a ﬁeld F has no ideals except {0} and F . For if a is a nonzero member of the ideal I, then ab = 1 for some b ∈ F , hence 1 ∈ I, and therefore I = F . Taking I to be the kernel of f , we see that I cannot be all of F because f (1) = 0. Thus I must be {0}, so that f is injective. ♣ 3.1.3 Theorem Let f be a nonconstant polynomial over the ﬁeld F . Then there is an extension E/F and an element α ∈ E such that f (α) = 0. Proof. Since f can be factored into irreducibles, we may assume without loss of generality that f itself is irreducible. The ideal I =< f (X) > in F [X] is prime (see (2.6.1)), in fact maximal (see (2.6.9)). Thus E = F [X]/I is a ﬁeld by (2.4.3). We have a problem at this point because F need not be a subset of E, but we can place an isomorphic copy of F inside E via the homomorphism h : a → a + I; by (3.1.2), h is a monomorphism, so we may identify F with a subﬁeld of E. Now let α = X + I; if f (X) = a0 + a1 X + · · · + an X n , then f (α) = (a0 + I) + a1 (X + I) + · · · + an (X + I)n = (a0 + a1 X + · · · + an X n ) + I = f (X) + I which is zero in E. ♣ The extension E is sometimes said to be obtained from F by adjoining a root α of f . Here is a further connection between roots and extensions. 3.1.4 Proposition Let f and g be polynomials over the ﬁeld F . Then f and g are relatively prime if and only if f and g have no common root in any extension of F . Proof. If f and g are relatively prime, their greatest common divisor is 1, so there are polynomials a(X) and b(X) over F such that a(X)f (X) + b(X)g(X) = 1. If α is a common

page 2 of Chapter 3

root of f and g, then the substitution of α for X yields 0 = 1, a contradiction. Conversely, if the greatest common divisor d(X) of f (X) and g(X) is nonconstant, let E be an extension of F in which d(X) has a root α (E exists by (3.1.3)). Since d(X) divides both f (X) and g(X), α is a common root of f and g in E. ♣ 3.1.5 Corollary If f and g are distinct monic irreducible polynomials over F , then f and g have no common roots in any extension of F . Proof. If h is a nonconstant divisor of the irreducible polynomials f and g, then up to multiplication by constants, h coincides with both f and g, so that f is a constant multiple of g. This is impossible because f and g are monic and distinct. Thus f and g are relatively prime, and the result follows from (3.1.4). ♣ If E is an extension of F and α ∈ E is a root of a polynomial f ∈ F [X], it is often of interest to examine the ﬁeld F (α) generated by F and α, in other words the smallest subﬁeld of E containing F and α (more precisely, containing all elements of F along with α). The ﬁeld F (α) can be described abstractly as the intersection of all subﬁelds of E containing F and α, and more concretely as the collection of all rational functions a0 + a1 α + · · · + am αm b 0 + b 1 α + · · · + bn α n with ai , bj ∈ F, m, n = 0, 1, . . ., and b0 + b1 α + · · · + bn αn = 0. In fact there is a much less complicated description of F (α), as we will see shortly. 3.1.6 Deﬁnitions and Comments If E is an extension of F , the element α ∈ E is said to be algebraic over F is there is a nonconstant polynomial f ∈ F [X] such that f (α) = 0; if α is not algebraic over F , it is said to be transcendental over F . If every element of E is algebraic over F , then E is said to be an algebraic extension of F . Suppose that α ∈ E is algebraic over F , and let I be the set of all polynomials g over F such that g(α) = 0. If g1 and g2 belong to I, so does g1 ± g2 , and if g ∈ I and c ∈ F [X], then cg ∈ I. Thus I is an ideal of F [X], and since F [X] is a PID, I consists of all multiples of some m(X) ∈ F [X]. Any two such generators must be multiples of each other, so if we require that m(X) be monic, then m(X) is unique. The polynomial m(X) has the following properties: (1) If g ∈ F [X], then g(α) = 0 if and only if m(X) divides g(X). This follows because g(α) = 0 iﬀ g(X) ∈ I, and I =< m(X) >, the ideal generated by m(X). (2) m(X) is the monic polynomial of least degree such that m(α) = 0. This follows from (1). (3) m(X) is the unique monic irreducible polynomial such that m(α) = 0. If m(X) = h(X)k(X) with deg h and deg k less than deg m, then either h(α) = 0 or k(α) = 0, so that by (1), either h(X) or k(X) is a multiple of m(X), which is impossible. Thus m(X) is irreducible, and uniqueness of m(X) follows from (3.1.5). The polynomial m(X) is called the minimal polynomial of α over F , sometimes written as min(α, F ). 3.1.7 Theorem If α ∈ E is algebraic over F and the minimal polynomial m(X) of α over F has degree n, then F (α) = F [α], the set of polynomials in α with coeﬃcients in F . In fact, F [α] is the set Fn−1 [α] of all polynomials of degree at most n − 1 with coeﬃcients in F , and 1, α, . . . , αn−1 form a basis for the vector space F [α] over the ﬁeld F . Consequently, [F (α) : F ] = n.

page 3 of Chapter 3

Proof. Let f (X) be any nonzero polynomial over F of degree n − 1 or less. Then since m(X) is irreducible and deg f < deg m, f (X) and m(X) are relatively prime, and there are polynomials a(X) and b(X) over F such that a(X)f (X) + b(X)m(X) = 1. But then a(α)f (α) = 1, so that any nonzero element of Fn−1 [α] has a multiplicative inverse. It follows that Fn−1 [α] is a ﬁeld. (This may not be obvious, since the product of two polynomials of degree n − 1 or less can have degree greater than n − 1, but if deg g > n − 1, then divide g by m to get g(X) = q(X)m(X) + r(X) where deg r(X) < deg m(X) = n. Replace X by α to get g(α) = r(α) ∈ Fn−1 [α]. Less abstractly, if m(α) = α3 + α + 1 = 0, then α3 = −α − 1, α4 = −α2 − α, and so on.) Now any ﬁeld containing F and α must contain all polynomials in α, in particular all polynomials of degree at most n − 1. Therefore Fn−1 [α] ⊆ F [α] ⊆ F (α). But F (α) is the smallest ﬁeld containing F and α, so F (α) ⊆ Fn−1 [α], and we conclude that F (α) = F [α] = Fn−1 [α]. Finally, the elements 1, α, . . . , αn−1 certainly span Fn−1 [α], and they are linearly independent because if a nontrivial linear combination of these elements were zero, we would have a nonzero polynomial of degree less than that of m(X) with α as a root, contradicting (2) of (3.1.6). ♣ We now prove a basic multiplicativity result for extensions, after a preliminary discussion. 3.1.8 Lemma Suppose that F ≤ K ≤ E, the elements αi , i ∈ I, form a basis for E over K, and the elements βj , j ∈ J, form a basis for K over F . (I and J need not be ﬁnite.) Then the products αi βj , i ∈ I, j ∈ J, form a basis for E over F . Proof. If γ ∈ E, then γ is a linear combination of the αi with coeﬃcients ai ∈ K, and each ai is a linear combination of the βj with coeﬃcients bij ∈ F . It follows that the αi βj span E over F . Now if i,j λij αi βj = 0, then i λij αi = 0 for all j, and consequently λij = 0 for all i, j, and the αi βj are linearly independent. ♣ 3.1.9 The Degree is Multiplicative If F ≤ K ≤ E, then [E : F ] = [E : K][K : F ]. In particular, [E : F ] is ﬁnite if and only if [E : K] and [K : F ] are both ﬁnite. Proof. In (3.1.8), we have [E : K] = |I|, [K : F ] = |J|, and [E : F ] = |I||J|. ♣ We close this section by proving that every ﬁnite extension is algebraic. 3.1.10 Theorem If E is a ﬁnite extension of F , then E is an algebraic extension of F . Proof. Let α ∈ E, and let n = [E : F ]. Then 1, α, α2 , . . . , αn are n + 1 vectors in an ndimensional vector space, so they must be linearly dependent. Thus α is a root of a nonzero polynomial with coeﬃcients in F , which means that α is algebraic over F . ♣ Problems For Section 3.1 1. Let E be an extension of F , and let S be a subset of E. If F (S) is the subﬁeld of E generated by S over F , in other words, the smallest subﬁeld of E containing F and S, describe F (S) explicitly, and justify your characterization. 2. If for each i ∈ I, Ki is a subﬁeld of the ﬁeld E, the composite of the Ki (notation i Ki ) is the smallest subﬁeld of E containing every Ki . As in Problem 1, describe the composite explicitly. 3. Assume that α is algebraic over F , with [F [α] : F ] = n. If β ∈ F [α], show that [F [β] : F ] ≤ n, in fact F [β] divides √ n.. 2 4. √ The minimal polynomial of 2 over the rationals √ Q is X − 2, by (3) of (3.1.6). Thus Q[ 2] consists of all numbers of√the form a0 + a1 2, where a0 and a1 are rational. By Problem 3, we know that −1 + 2 has a minimal polynomial over Q of degree at most 2. Find this minimal polynomial. 5. If α is algebraic over F and β belongs to F [α], describe a systematic procedure for ﬁnding the minimal polynomial of β over F .

page 4 of Chapter 3

6. If E/F and the element α ∈ E is transcendental over F , show that F (α) is isomorphic to F (X), the ﬁeld of rational functions with coeﬃcients in F . 7. Theorem 3.1.3 gives one method of adjoining a root of a polynomial, and in fact there is essentially only one way to do this. If E is an extension of F and α ∈ E is algebraic over F with minimal polynomial m(X), let I be the ideal < m(X) >⊆ F [X]. Show that F (α) is isomorphic to F [X]/I. [Deﬁne ϕ : F [X] → E by ϕ(f (X)) = f (α), and use the ﬁrst isomorphism theorem for rings.] 8. In the proof of (3.1.3), we showed that if f is irreducible in F [X], then I =< f > is a maximal ideal. Show that conversely, if I is a maximal ideal, then f is irreducible. 9. Suppose that F ≤ E ≤ L, with α ∈ L. What is the relation between the minimal polynomial of α over F and the minimal polynomial of α over E? 10. If α1 , . . . , αn are algebraic over F , we can successively adjoin the αi to F to obtain the ﬁeld F [α1 , . . . , αn ] consisting of all polynomials over F in the αi . Show that [F [α1 , . . . , αn ] : F ] ≤

n

[F (αi ) : F ] < ∞

i=1

3.2 Splitting Fields If f is a polynomial over the ﬁeld F , then by (3.1.3) we can ﬁnd an extension E1 of F containing a root α1 of f . If not all roots of f lie in E1 , we can ﬁnd an extension E2 of E1 containing another root α2 of f . If we continue the process, eventually we reach a complete factorization of f . In this section we examine this idea in detail. If E is an extension of F and α1 , . . . , αk ∈ E, we will use the notation F (α1 , . . . , αk ) for the subﬁeld of E generated by F and the αi . Thus F (α1 , . . . , αk ) is the smallest subﬁeld of E (that is, the intersection of all subﬁelds of E) containing all elements of F along with the αi . Explicitly, F (α1 , . . . , αk ) is the collection of all rational functions in the αi with nonzero denominators. 3.2.1 Deﬁnitions and Comments If E is an extension of F and f ∈ F [X], we say that f splits over E if f can be written as λ(X − α1 ) · · · (X − αk ) for some α1 , . . . , αk ∈ E and λ ∈ F. [There is a subtle point that should be mentioned. We would like to refer to the αi as “the” roots of f , but in doing so we are implicitly assuming that if β is an element of some extension E of E and f (β) = 0, then β must be one of the αi . This follows upon substituting β into the equation f (X) = λ(X − α1 ) · · · (X − αk ) = 0.] If K is an extension of F and f ∈ F [X], we say that K is a splitting ﬁeld for f over F if f splits over K but not over any proper subﬁeld of K containing F . Equivalently, K is a splitting ﬁeld for f over F if f splits over K and K is generated over F by the roots α1 , . . . , αk of f , in other words, F (α1 , . . . , αk ) = K. For if K is a splitting ﬁeld for f , then since f splits over K we have all αj ∈ K, so F (α1 , . . . , αk ) ⊆ K. But f splits over F (α1 , . . . , αk ), and it follows that F (α1 , . . . , αk ) cannot be a proper subﬁeld; it must coincide with K. Conversely, if f splits over K and F (α1 , . . . , αk ) = K, let L be a subﬁeld of K containing F . If f splits over L then all αi belong to L, so K = F (α1 , . . . , αk ) ⊆ L ⊆ K, so L = K. If f ∈ F [X] and f splits over the extension E of F , then E contains a unique splitting ﬁeld for f , namely F (α1 , . . . , αk ). 3.2.2 Proposition If f ∈ F [X] and deg f = n, then f has a splitting ﬁeld K over F with [K : F ] ≤ n!. Proof. By (3.1.3), F has an extension E1 containing a root α1 of f , and the extension F (α1 )/F has degree at most n. (Since f (α1 ) = 0, the minimal polynomial of α1 divides f ;

page 5 of Chapter 3

see (3.1.6) and (3.1.7).) We may then write f (X) = λ(X −α1 )r1 g(X), where α1 is not a root of g and deg g ≤ n − 1. If g is nonconstant, we can ﬁnd an extension of F (α1 ) containing a root α2 of g, and the extension F (α1 , α2 ) will have degree at most n − 1 over F (α1 ). Continue inductively and use (3.1.9) to reach an extension of degree at most n! containing all the roots of f . ♣ If f ∈ F [X] and f splits over E, then we may pick any root α of f and adjoin it to F to obtain the extension F (α). Roots of the same irreducible factor of f yield essentially the same extension, as the next result shows. 3.2.3 Theorem If α and β are roots of the irreducible polynomial f ∈ F [X] in an extension E of F , then F (α) is isomorphic to F (β) via an isomorphism that carries α into β and is the identity on F . Proof. Without loss of generality we may assume f monic (if not, divide f by its leading coeﬃcient). By (3.1.6), part (3), f is the minimal polynomial of both α and β. By (3.1.7), the elements of F (α) can be expressed uniquely as a0 + a1 α + · · · + an−1 αn−1 , where the ai belong to F and n is the degree of f . The desired isomorphism is given by a0 + a1 α + · · · + an−1 αn−1 → a0 + a1 β + · · · + an−1 β n−1 . ♣ If f is a polynomial in F [X] and F is isomorphic to the ﬁeld F via the isomorphism i, we may regard f as a polynomial over F . We simply use i to transfer f . Thus if f = a0 + a1 X + · · · an X n , then f = i(f ) = i(a0 ) + i(a1 )X + · · · + i(an )X n . There is only a notational diﬀerence between f and f , and we expect that splitting ﬁelds for f and f should also be essentially the same. We prove this after the following deﬁnition. 3.2.4 Deﬁnition If E and E are extensions of F and i is an isomorphism of E and E , we say that i is an F -isomorphism if i ﬁxes F , that is, i(a) = a for every a ∈ F . F homomorphisms, F -monomorphisms, etc., are deﬁned similarly. 3.2.5 Isomorphism Extension Theorem Suppose that F and F are isomorphic, and the isomorphism i carries the polynomial f ∈ F [X] to f ∈ F [X]. If K is a splitting ﬁeld for f over F and K is a splitting ﬁeld for f over F , then i can be extended to an F -isomorphism of K and K . In particular, if F = F and i is the identity function, we conclude that any two splitting ﬁelds of f are F -isomorphic. Proof. Carry out the construction of a splitting ﬁeld for f over F as in (3.2.2), and perform exactly the same steps to construct a splitting ﬁeld for f over F . At every stage, there is only a notational diﬀerence between the ﬁelds obtained. Furthermore, we can do the ﬁrst construction inside K and the second inside K . But the comment at the end of (3.2.1) shows that the splitting ﬁelds that we have constructed coincide with K and K . ♣ 3.2.6 Example Find a splitting ﬁeld for f (X) = X 3 − 2 over the rationals Q.

√ If α is the positive cube root of 2, then the roots of f are α, α(− 12 + i 12 3) and √ α(− 12 − i 12 3). The polynomial f is irreducible, either by Eisenstein’s criterion or by the observation hat if f were factorable, it would have a linear factor, and there is no rational number whose√cube is 2. Thus f is the minimal polynomial of α, so [Q(α) : Q] √ = 3. Now since α and i 3 generate all the roots of f , the splitting ﬁeld is K = Q(α, i√ 3). (We regard all ﬁelds in this example as subﬁelds of the i 3∈ / Q(α) √ complex numbers C.) Since √ (because Q(α) is a subﬁeld of the reals),[Q(α, √ i 3) : Q(α)] is at least 2. But i 3 is a root of X 2 + 3 ∈ Q(α)[X], so the degree of Q(α, i 3) over Q(α) is a most 2, and therefore is exactly 2. Thus √ √ [K : Q] = [Q(α, i 3) : Q] = [Q(α, i 3) : Q(α)][Q(α) : Q] = 2 × 3 = 6. Problems For Section 3.2

page 6 of Chapter 3

1. Find a splitting ﬁeld for f (X) = X 2 − 4X + 4 over Q. 2. Find a splitting ﬁeld K for f (X) = X 2 − 2X + 4 over Q, and determine the degree of K over Q. 3. Find a splitting ﬁeld K for f (X) = X 4 − 2 over Q, and determine [K : Q]. 4. Let C be a family of polynomials over f , and let K be an extension of f . Show that the following two conditions are equivalent: (a) Each f ∈ C splits over K, but if F ≤ K < K, then it is not true that each f ∈ C splits over K . (b) Each f ∈ C splits over K, and K is generated over F by the roots of all the polynomials in C. If one, and hence both, of these conditions are satisﬁed, we say that K is a splitting ﬁeld for C over F . 5. Suppose that K is a splitting ﬁeld for the ﬁnite set of polynomials {f1 , . . . , fr } over F . Express K as a splitting ﬁeld for a single polynomial f over F . 6. √If m √ and n are distinct square-free positive integers, show that the splitting ﬁeld Q( m, n) of (X 2 − m)(X 2 − n) has degree 4 over Q. 3.3 Algebraic Closures If f is a polynomial of degree n over the rationals or the reals, or more generally over the complex numbers, then f need not have any rational roots, or even real roots, but we know that f always has n complex roots, counting multiplicity. This favorable situation can be duplicated for any ﬁeld F , that is, we can construct an algebraic extension C of F with the property that any polynomial in C[X] splits over C. There are many ways to express this idea. 3.3.1 Proposition If C is a ﬁeld, the following conditions are equivalent. (1) (2) (3) (4)

Every Every Every C has

nonconstant polynomial f ∈ C[X] has at least one root in C. nonconstant polynomial f ∈ C[X] splits over C. irreducible polynomial f ∈ C[X] is linear. no proper algebraic extensions.

If any (and hence all) of these conditions are satisﬁed, we say that C is algebraically closed. Proof. (1) implies (2): By (1) we may write f = (X − α1 )g. Proceed inductively to show that any nonconstant polynomial is a product of linear factors. (2) implies (3): If f is an irreducible polynomial in C[X], then by (2.9.1), f is nonconstant. By (2), f is a product of linear factors. But f is irreducible, so there can be only one such factor. (3) implies (4): Let E be an algebraic extension of C. If α ∈ E, let f be the minimal polynomial of α over C. Then f is irreducible and by (3), f is of the form X − α. But then α ∈ C, so E = C. (4) implies (1): Let f be a nonconstant polynomial in C[X], and adjoin a root α of f to obtain C(α), as in (3.1.3). But then C(α) is an algebraic extension of C, so by (4), α ∈ C. ♣ It will be useful to embed an arbitrary ﬁeld F in an algebraically closed ﬁeld. 3.3.2 Deﬁnitions and Comments An extension C of F is an algebraic closure of F if C is algebraic over F and C is algebraically closed. Note that C is minimal among algebraically closed extensions of F . For if F ≤ K ≤ C and α ∈ C, α ∈ / K, then since α is algebraic over F it is algebraic over K. But since α ∈ / K,

page 7 of Chapter 3

the minimal polynomial of α over K is a nonlinear irreducible polynomial in K[X]. By (3) of (3.3.1), K cannot be algebraically closed. If C is an algebraic extension of F , then in order for C to be an algebraic closure of F it is suﬃcient that every polynomial in F [X] (rather than C[X]) splits over C. To prove this, we will need the following result. 3.3.3 Proposition If E is generated over F by ﬁnitely many elements α1 , . . . , αn algebraic over F (so that E = F (α1 , . . . , αn )), then E is a ﬁnite extension of F . Proof. Set E0 = F and Ek = F (α1 , . . . , αk ), 1 ≤ k ≤ n (so En = E). Then Ek = Ek−1 (αk ), the degree where αk is algebraic over F and hence over Ek−1 . But by (3.1.7), [Ek : Ek−1 ] is n of the minimal polynomial of αk over Ek−1 , which is ﬁnite. By (3.1.9), [E : F ] = k=1 [Ek : Ek−1 ] < ∞. ♣ 3.3.4 Corollary If E is an extension of F and A is the set of all elements in E that are algebraic over F (the algebraic closure of F in E), then A is a subﬁeld of E. Proof. If α, β ∈ A, then the sum, diﬀerence, product and quotient (if β = 0) of α and β belong to F (α, β), which is a ﬁnite extension of F by (3.3.3), and therefore an algebraic extension of F by (3.1.10). But then α + β, α − β, αβ and α/β belong to A, proving that A is a ﬁeld. ♣ 3.3.5 Corollary (Transitivity of Algebraic Extensions) If E is algebraic over K (in other words, every element of E is algebraic over K), and K is algebraic over F , then E is algebraic over F . Proof. Let α ∈ E, and let m(X) = b0 + b1 X + . . . + bn−1 X n−1 + X n be the minimal polynomial of α over K. The bi belong to K and are therefore algebraic over F . If L = F (b0 , b1 , . . . , bn−1 ), then by (3.3.3), L is a ﬁnite extension of F . Since the coeﬃcients of m(X) belong to L, α is algebraic over L, so by (3.1.7), L(α) is a ﬁnite extension of L. By (3.1.9), L(α) is a ﬁnite extension of F . By (3.1.10), α is algebraic over F . ♣ Now we can add another condition to (3.3.1). 3.3.6 Proposition Let C be an algebraic extension of F . Then C is an algebraic closure of F if and only if every nonconstant polynomial in F [X] splits over C. Proof. The ”only if” part follows from (2) of (3.3.1), since F ⊆ C. Thus assume that every nonconstant polynomial in F [X] splits over C. If f is a nonconstant polynomial in C[X], we will show that f has at least one root in C, and it will follow from (1) of (3.3.1) that C is algebraically closed. Adjoin a root α of f to obtain the extension C(α). Then C(α) is algebraic over C by (3.1.7), and C is algebraic over F by hypothesis. By (3.3.5), C(α) is algebraic over F , so α is algebraic over F . But then α is a root of some polynomial g ∈ F [X], and by hypothesis, g splits over C. By deﬁnition of ”splits” (see (3.2.1)), all roots of g lie in C, in particular α ∈ C. Thus f has at least one root in C. ♣ To avoid a lengthy excursion into formal set theory, we argue intuitively to establish the following three results. (For complete proofs, see Morandi, “Fields and Galois Theory”, pages 31,35; Roman, “Field Theory”, p.50.) 3.3.7 Theorem Every ﬁeld F has an algebraic closure. Informal argument. Well-order F [X] and use transﬁnite induction, beginning with the ﬁeld F0 = F . At stage f we adjoin all roots of the polynomial f by constructing a splitting ﬁeld for f over the ﬁeld F
page 8 of Chapter 3

Informal argument. Carry out the recursive procedure described in (3.3.7) in both C and C . At each stage we may use the fact that any two splitting ﬁelds of the same polynomial are F -isomorphic; see (3.2.5). When we ﬁnish, we have F -isomorphic algebraic closures of F , say D ⊆ C and D ⊆ C . But an algebraic closure is a minimal algebraically closed extension by (3.3.2), and therefore D = C and D = C . ♣ 3.3.9 Theorem If E is an algebraic extension of F , C is an algebraic closure of F , and i is an embedding (that is, a monomorphism) of F into C, then i can be extended to an embedding of E into C. Informal argument. Each α ∈ E is a root of some polynomial in F [X], so if we allow α to range over all of E, we get a collection S of polynomials in F [X]. Within C, carry out the recursive procedure of (3.3.7) on the polynomials in S. The resulting ﬁeld lies inside C and contains an F -isomorphic copy of E. ♣ Problems For Section 3.3 1. Show that the converse of (3.3.3) holds, that is, if E is a ﬁnite extension of F , then E is generated over F by ﬁnitely many elements that are algebraic over F . 2. An algebraic number is a complex number that is algebraic over the rational ﬁeld Q. A transcendental number is a complex number that is not algebraic over Q. Show that there only countably many algebraic numbers, and consequently there are uncountably many transcendental numbers. 3. Give an example of an extension C/F such that C is algebraically closed but C is not an algebraic extension of F . 4. Give an example of an extension E/F such that E is an algebraic but not a ﬁnite extension of F . 5. In the proof of (3.3.7), why is C algebraic over F ? 6. Show that the set A of algebraic numbers is an algebraic closure of Q. 7. If E is an algebraic extension of the inﬁnite ﬁeld F , show that |E| = |F |. 8. Show that any set S of nonconstant polynomials in F [X] has a splitting ﬁeld over F . 9. Show that an algebraically closed ﬁeld must be inﬁnite. 3.4 Separability If f is a polynomial in F [X], we can construct a splitting ﬁeld K for f over F , and all roots of f must lie in K. In this section we investigate the multiplicity of the roots. 3.4.1 Deﬁnitions and Comments An irreducible polynomial f ∈ F [X] is separable if f has no repeated roots in a splitting ﬁeld; otherwise f is inseparable. If f is an arbitrary polynomial, not necessarily irreducible, then we call f separable if each of its irreducible factors is separable. Thus if f (X) = (X − 1)2 (X − 3) over Q, then f is separable, because the irreducible factors (X − 1) and (X − 3) do not have repeated roots. We will see shortly that over a ﬁeld of characteristic 0 (for example, the rationals), every polynomial is separable. Here is a method for testing for multiple roots. 3.4.2 Proposition If f (X) = a0 + a1 X + · · · + an X n ∈ F [X], let f be the derivative of f , deﬁned by f (X) = a1 + 2a2 X + · · · + nan X n−1 . [Note that the derivative is a purely formal expression; we completely ignore questions about existence of limits. One can check by brute force that the usual rules for diﬀerentiating a sum and product apply]. If g is the greatest common divisor of f and f , then f has a repeated root in a splitting ﬁeld if and only if the degree of g is at least 1.

page 9 of Chapter 3

Proof. If f has a repeated root, we can write f (X) = (X − α)r h(X) where r ≥ 2. Applying the product rule for derivatives, we see that (X − α) is a factor of both f and f , and consequently deg g ≥ 1. Conversely, if deg g ≥ 1, let α be a root of g in some splitting ﬁeld. Then (X − α) is a factor of both f and f . We will show that α is a repeated root of f . If not, we may write f (X) = (X − α)h(X) where h(α) = 0. Diﬀerentiate to obtain f (X) = (X − α)h (X) + h(X), hence f (α) = h(α) = 0. This contradicts the fact that (X − α) is a factor of f . ♣ 3.4.3 Corollary (1) Over a ﬁeld of characteristic zero, every polynomial is separable. (2) Over a ﬁeld F of prime characteristic p, the irreducible polynomial f is inseparable if and only if f is the zero polynomial. Equivalently, f is a polynomial in X p ; we abbreviate this as f ∈ F [X p ]. Proof. (1) Without loss of generality, we can assume that we are testing an irreducible polynomial f . The derivative of X n is nX n−1 , and in a ﬁeld of characteristic 0, n cannot be 0. Thus f is a nonzero polynomial whose degree is less than that of f . Since f is irreducible, the gcd of f and f is either 1 or f , and the latter is excluded because f cannot possibly divide f . By (3.4.2), f is separable. (2) If f = 0, the argument of (1) shows that f is separable. If f = 0, then gcd(f, f ) = f , so by (3.4.2), f is inseparable. In characteristic p, an integer n is zero if and only if n is a multiple of p, and it follows that f = 0 iﬀ f ∈ F [X p ]. ♣ By (3.4.3), part (1), every polynomial over the rationals (or the reals or the complex numbers) is separable. This pleasant property is shared by ﬁnite ﬁelds as well. First note that a ﬁnite ﬁeld F cannot have characteristic 0, since a ﬁeld of characteristic 0 must contain a copy of the integers (and the rationals as well), and we cannot squeeze inﬁnitely many integers into a ﬁnite set. Now recall is simply the binomial expansion modulo p, which n n n (a + b)p = ap + bp , since p divides kp for 1 ≤ k ≤ p − 1. [By induction, (a + b)p = ap + bp for every positive integer n.] Here is the key step in the analysis. 3.4.4 The Frobenius Automorphism Let F be a ﬁnite ﬁeld of characteristic p, and deﬁne f : F → F by f (α) = αp . Then f is an automorphism. In particular, if α ∈ F then α = β p for some β ∈ F . Proof. We have f (1) = 1 and f (α + β) = (α + β)p = αp + β p = f (α) + f (β) and f (αβ) = (αβ)p = αp β p = f (α)f (β) so f is a monomorphism. But an injective function from a ﬁnite set to itself is automatically surjective, and the result follows. ♣ 3.4.5 Proposition Over a ﬁnite ﬁeld, every polynomial is separable. Proof. Suppose that f is an irreducible polynomial over the ﬁnite ﬁeld F with repeated roots in a splitting ﬁeld. By (3.4.3), part (2), f (X) has the form a0 + a1 X p + · · · + an X np with the ai ∈ F . By (3.4.4), for each i there is an element bi ∈ F such that bpi = ai . But then (b0 + b1 X + · · · + bn X n )p = bp0 + bp1 X p + · · · + bn X np = f (X) which contradicts the irreducibility of f . ♣ Separability of an element can be deﬁned in terms of its minimal polynomial. 3.4.6 Deﬁnitions and Comments If E is an extension of F and α ∈ E, then α is separable over F if α is algebraic over F and min(α, F ) is a separable polynomial. If every element

page 10 of Chapter 3

of E is separable over F , we say that E is a separable extension of F or the extension E/F is separable or E is separable over F . By (3.4.3) and (3.4.5), every algebraic extension of a ﬁeld of characteristic zero or a ﬁnite ﬁeld is separable. 3.4.7 Lemma If F ≤ K ≤ E and E is separable over F , then K is separable over F and E is separable over K. Proof. Since K is a subﬁeld of E, K/F is separable. If α ∈ E, then since α is a root of min(α, F ), it follows from (1) of (3.1.6) that min(α, K) divides min(α, F ). By hypothesis, min(α, F ) has no repeated roots in a splitting ﬁeld, so neither does min(α, K). Thus E/K is separable. ♣ The converse of (3.4.7) is also true: If K/F and E/K are separable, then E/F is separable. Thus we have transitivity of separable extensions. We will prove this (for ﬁnite extensions) in the exercises. In view of (3.4.6), we can produce many examples of separable extensions. Inseparable extensions are less common, but here is one way to construct them. 3.4.8 Example Let F = Fp (t) be the set of rational functions (in the indeterminate t) with coeﬃcients in the ﬁeld with p elements (the integers mod p). Thus an element of F looks like a0 + a1 t + · · · + am tm . b 0 + b 1 t + · · · + bn tn √ with the ai and bj in Fp . Adjoin p t, that is, a root of X p − t, to create the extension E. Note that X p − t is irreducible by Eisenstein, because t is irreducible in Fp [t]. (The product of two nonconstant polynomials in t cannot possibly be t.) The extension E/F is inseparable, since √ √ p p X p − t = X p − ( t)p = (X − t)p , which has multiple roots. Problems For Section 3.4 1. Give an example of a separable polynomial f whose derivative is zero. (In view of (3.4.3), f cannot be irreducible.) 2. Let α ∈ E, where E is an algebraic extension of a ﬁeld F of prime characteristic p. Let m(X) be the minimal polynomial of α over the ﬁeld F (αp ). Show that m(X) splits over E, and in fact α is the only root, so that m(X) is a power of (X − α). 3. Continuing Problem 2, if α is separable over the ﬁeld F (αp ), show that α ∈ F (αp ). 4. A ﬁeld F is said to be perfect if every polynomial over F is separable. Equivalently, every algebraic extension of F is separable. Thus ﬁelds of characteristic zero and ﬁnite ﬁelds are perfect. Show that if F has prime characteristic p, then F is perfect if and only if every element of F is the pth power of some element of F . For short we write F = F p . We now turn to transitivity of separable extensions. 5. Let E be a ﬁnite extension of a ﬁeld F of prime characteristic p, and let K = F (E p ) be the subﬁeld of E obtained from F by adjoining the pth powers of all elements of E. Show that F (E p ) consists of all ﬁnite linear combinations of elements in E p with coeﬃcients in F. 6. Let E be a ﬁnite extension of the ﬁeld F of prime characteristic p, and assume that E = F (E p ). If the elements y1 , . . . , yr ∈ E are linearly independent over F , show that y1p , . . . , yrp are linearly independent over F . 7. Let E be a ﬁnite extension of the ﬁeld F of prime characteristic p. Show that the extension is separable if and only if E = F (E p ). 8. If F ≤ K ≤ E with [E : F ] < ∞, with E separable over K and K separable over F , show that E is separable over F . 9. Let f be an irreducible polynomial in F [X], where F has characteristic p > 0. Express

page 11 of Chapter 3 m

f (X) as g(X p ), where the nonnegative integer m is a large as possible. (This makes sense 0 because X p = X, so m = 0 always works, and f has ﬁnite degree, so m is bounded above.) Show that g is irreducible and separable. m 10. Continuing Problem 9, if f has only one distinct root α, show that αp ∈ F . 11. If E/F , where char F = p > 0, and the element α ∈ E is algebraic over F , show that n the minimal polynomial of α over F has only one distinct root if and only if αp ∈ F for some nonnegative integer n. (In this case we say that α is purely inseparable over F .) 3.5 Normal Extensions Let E/F be a ﬁeld extension. In preparation for Galois theory, we are going to look at monomorphisms deﬁned on E, especially those which ﬁx F . First we examine what an F -monomorphism does to the roots of a polynomial in F [X]. 3.5.1 Lemma Let σ : E → E be an F -monomorphism, and assume that the polynomial f ∈ F [X] splits over E. If α is a root of f in E, then so is σ(α). Thus σ permutes the roots of f . Proof. If b0 + b1 α + · · · + bn αn = 0, with the bi ∈ F , apply σ and note that since σ is an F -monomorphism, σ(bi ) = bi and σ(αi ) = (σ(α))i . Thus b0 + b1 σ(α) + · · · + bn (σ(α))n = 0. ♣ Now let C be an algebraic closure of E. It is convenient to have C available because it will contain all the roots of a polynomial f ∈ E[X], even if f does not split over E. We are going to count the number of embeddings of E in C that ﬁx F , that is, the number of F -monomorphisms of E into C. Here is the key result. 3.5.2 Theorem Let E/F be a ﬁnite separable extension of degree n, and let σ be an embedding of F in C. Then σ extends to exactly n embeddings of E in C, in other words, there are exactly n embeddings τ of E in C such that the restriction τ |F of τ to F coincides with σ. In particular, taking σ to be the identity function on F , there are exactly n F monomorphisms of E into C. Proof. An induction argument works well. If n = 1 then E = F and there is nothing to prove, so assume n > 1 and choose an element α that belongs to E but not to F . If f is the minimal polynomial of α over F , let g = σ(f ). (This is a useful shorthand notation, indicating that if ai is one of the coeﬃcients of f , the corresponding coeﬃcient of g is σ(ai ).) Any factorization of g can be translated via the inverse of σ to a factorization of f , so g is separable and irreducible over the ﬁeld σ(F ). If β is any root of g, then there is a unique isomorphism of F (α) and (σ(F ))(β) that carries α into β and coincides with σ on F . Explicitly, b0 + b1 α + · · · + br αr → σ(b0 ) + σ(b1 )β + · · · + σ(br )β r . Now if deg g = r, then [F (α) : F ] = deg f = deg g = r as well, so by (3.1.9), [E : F (α)] = n/r < n. By separability, g has exactly r distinct roots in C, so there are exactly r possible choices of β. In each case, the resulting embedding of F (α) in C has, by the induction hypothesis, exactly n/r extensions to embeddings of E in C. This produces n distinct embeddings of E in C extending σ. But if τ is any embedding of F in C that extends σ, then just as in (3.5.1), τ must take α to a root of g, i.e., to one of the β’s. If there were more than n possible τ ’s, there would have to be more than n/r possible extensions of at least one of the embeddings of F (α) in C. This would contradict the induction hypothesis. ♣ √ 3.5.3 Example Adjoin the positive cube root of 2 to the rationals to get E √ = Q( 3 2). √ √ The roots of the irreducible polynomial f (X) = X√3 − 2 are 3 2, ω 3 2 and ω 2 3 2, where √ 1 1 i2π/3 ω=e = − 2 + 2 3 and ω 2 = ei4π/3 = − 12 − 12 3.

page 12 of Chapter 3

Notice that the polynomial f has a root in E but does not split in E (because the other two roots are complex and E consists entirely of real numbers). We give a special name to extensions that do not have this annoying drawback. 3.5.4 Deﬁnition The algebraic extension E/F is normal (we also say that E is normal over F ) if every irreducible polynomial over F that has at least one root in E splits over E. In other words, if α ∈ E, then all conjugates of α over F (i.e., all roots of the minimal polynomial of α over F ) belong to E. Here is an equivalent condition. (The hypothesis that E/F is ﬁnite rather than simply algebraic can be removed, but we will not need the more general result.) 3.5.5 Theorem The ﬁnite extension E/F is normal if and only if every F -monomorphism of E into an algebraic closure C is actually an F -automorphism of E. Proof. If E/F is normal, then as in (3.5.1), an F -monomorphism τ of E into C must map each element of E to one of its conjugates. Thus by hypothesis, τ (E) ⊆ E. But τ (E) is an isomorphic copy of E , so it must have the same degree as E over F . Since the degree is assumed ﬁnite, we have τ (E) = E. (All we are saying here is that an m-dimensional subspace of an m-dimensional vector space is the entire space.) Conversely, let α ∈ E, and let β be any conjugate of α over F . As in the proof of (3.5.2), there is an F -monomorphism of E into C that carries α to β. If all such embeddings are F -automorphisms of E, we must have β ∈ E, and we conclude that E is normal over F . ♣ 3.5.6 Remarks In (3.5.2) and (3.5.5), the algebraic closure can be replaced by any ﬁxed normal extension of F containing E; the proof is the same. Also, the implication τ (E) ⊆ E ⇒ τ (E) = E holds for any F -monomorphism τ and any ﬁnite extension E/F ; normality is not involved. The next result yields many explicit examples of normal extensions. 3.5.7 Theorem The ﬁnite extension E/F is normal if and only if E is a splitting ﬁeld for some polynomial f ∈ F [X]. Proof. Assume that E is normal over F . Let α1 , . . . , αn be a basis for E over F , and let fi be the minimal polynomial of αi over F, i = 1, . . . , n. Since fi has a root αi in E, fi splits over E, hence so does f = f1 · · · fn . If f splits over a ﬁeld K with F ⊆ K ⊆ E, then each αi belongs to K, and therefore K must coincide with E. Thus E is a splitting ﬁeld for f over F . Conversely, let E be a splitting ﬁeld for f over F , where the roots of f are αi , i = 1, . . . , n. Let τ be an F -monomorphism of E into an algebraic closure. As in (3.5.1), τ takes each αi into another root of f , and therefore τ takes a polynomial in the αi to another polynomial in the αi . But F (α1 , . . . , αn ) = E, so τ (E) ⊆ E. By (3.5.6), τ is an automorphism of E, so by (3.5.5), E/F is normal. ♣ 3.5.8 Corollary Let F ≤ K ≤ E, where E is a ﬁnite extension of F . If E/F is normal, so is E/K. Proof. By (3.5.7), E is a splitting ﬁeld for some polynomial f ∈ F [X], so that E is generated over F by the roots of f . But then f ∈ K[X] and E is generated over K by the roots of f . Again by (3.5.7), E/K is normal. ♣ 3.5.9 Deﬁnitions and Comments If E/F is normal and separable, it is said to be a Galois extension; we also say that E is Galois over F . It follows from (3.5.2) and (3.5.5) that if E/F is a ﬁnite Galois extension, then there are exactly [E : F ] F -automorphisms of E. If E/F is ﬁnite and separable but not normal, then at least one F -embedding of E into an algebraic closure must fail to be an automorphism of E. Thus in this case, the number of F -automorphisms of E is less than the degree of the extension. If E/F is an arbitrary extension, the Galois group of the extension, denoted by Gal(E/F ), is the set of F -automorphisms of E. (The set is a group under composition of functions.)

page 13 of Chapter 3

√ 3.5.10 Example Let E = Q( 3 2), as in (3.5.3). The Galois group of the extension √ consists of the identity automorphism alone. For any Q-monomorphism σ of E must take√3 2 into a root of X 3 − 2. Since the other two roots are complex and do√not belong to E, 3 2 must map to itself. But σ is completely determined by its action on 3 2, and the result follows. If E/F is not normal, we can always enlarge E to produce a normal extension of F . If C is an algebraic closure of E, then C contains all the roots of every polynomial in F [X], so C/F is normal. Let us try to look for a smaller normal extension. 3.5.11 The Normal Closure Let E be a ﬁnite extension of F , say E = F (α1 , . . . , αn ). If N ⊇ E is any normal extension of F , then N must contain the αi along with all conjugates of the αi , that is, all roots of min(αi , F ), i = 1, . . . , n. Thus if f is the product of these minimal polynomials, then N must contain the splitting ﬁeld K for f over F . But K/F is normal by (3.5.7), so K must be the smallest normal extension of F that contains E. It is called the normal closure of E over F . We close the section with an important result on the structure of ﬁnite separable extensions. 3.5.12 Theorem of the Primitive Element If E/F is a ﬁnite separable extension, then E = F (α) for some α ∈ E. We say that α is a primitive element of E over F . Proof. We will argue by induction on n = [E : F ]. If n = 1 then E = F and we can take α to be any member of F . If n > 1, choose α ∈ E \ F . By the induction hypothesis, there is a primitive element β for E over F (α), so that E = F (α, β). We are going to show that if c ∈ F is properly chosen, then E = F (α + cβ). Now by (3.5.2), there are exactly n F -monomorphisms of E into an algebraic closure C, and each of these maps restricts to an F -monomorphism of F (α + cβ) into C. If F (α + cβ) = E, then [F (α + cβ) : F ] < n, and it follows from (3.5.2) that at least two embeddings of E, say σ and τ , must coincide when restricted. Therefore σ(α) + cσ(β) = τ (α) + cτ (β), hence

σ(α) − τ (α) (1) τ (β) − σ(β) (If τ (β) = σ(β) then by the previous equation, τ (α) = σ(α). But an F -embedding of E is determined by what it does to α and β, hence σ = τ , a contradiction.) Now an F monomorphism must map α to one of its conjugates over F , and similarly for β. Thus there are only ﬁnitely many possible values for the ratio in (1). If we select c to be diﬀerent from each of these values, we reach a contradiction of our assumption that F (α + cβ) = E. The proof is complete if F is an inﬁnite ﬁeld. We must leave a gap here, to be ﬁlled later (see (6.4.4)). If F is ﬁnite, then so is E (since E is a ﬁnite-dimensional vector space over F ). We will show that the multiplicative group of nonzero elements of a ﬁnite ﬁeld E is cyclic, so if α is a generator of this group, then E = F (α). ♣ c=

Problems For Section 3.5 1. Give an example √ of ﬁelds F ≤ K ≤ E such that E/F is normal but K/F is not. 2. Let E = Q( a), where a is an integer that is not a perfect square. Show that E/Q is normal. 3. Give an example of ﬁelds F ≤ K ≤ E such that E/K and K/F are normal, but E/F is not. Thus transitivity fails for normal extensions. 4. Suppose that in (3.5.2), the hypothesis of separability is dropped. State and prove an appropriate conclusion. √ √ 5. Show that E = Q( 2, 3) is a Galois extension of Q. 6. In Problem 5, ﬁnd the Galois group of E/Q. 7. Let E be a ﬁnite extension of F , and let K be a normal closure (= minimal normal extension) of E over F , as in (3.5.11). Is K unique? 8. If E1 and E2 are normal extensions of F , show that E1 ∩ E2 is normal over F .

page 1 of Chapter 4

CHAPTER 4 MODULE FUNDAMENTALS 4.1 Modules and Algebras 4.1.1 Deﬁnitions and Comments A vector space M over a ﬁeld R is a set of objects called vectors, which can be added, subtracted and multiplied by scalars (members of the underlying ﬁeld). Thus M is an abelian group under addition, and for each r ∈ R and x ∈ M we have an element rx ∈ M . Scalar multiplication is distributive and associative, and the multiplicative identity of the ﬁeld acts as an identity on vectors. Formally, r(x + y) = rx + ry; (r + s)x = rx + sx; r(sx) = (rs)x; 1x = x for all x, y ∈ M and r, s ∈ R. A module is just a vector space over a ring. The formal deﬁnition is exactly as above, but we relax the requirement that R be a ﬁeld, and instead allow an arbitrary ring. We have written the product rx with the scalar r on the left, and technically we get a left R-module over the ring R. The axioms of a right R-module are (x + y)r = xr + yr; x(r + s) = xr + xs; (xs)r = x(sr), x1 = x. “Module” will always mean left module unless stated otherwise. Most of the time, there is no reason to switch the scalars from one side to the other (especially if the underlying ring is commutative). But there are cases where we must be very careful to distinguish between left and right modules (see Example 6 of (4.1.3)). 4.1.2 Some Basic Properties of Modules Let M be an R-module. The technique given for rings in (2.1.1) can be applied to establish the following results, which hold for any x ∈ M and r ∈ R. We distinguish the zero vector 0M from the zero scalar 0R . (1) r0M = 0M [r0M = r(0M + 0M ) = r0M + r0M ] (2) 0R x = 0M [0R x = (0R + 0R )x = 0R x + 0R x] (3) (−r)x = r(−x) = −(rx) [as in (2) of (2.1.1) with a replaced by r and b by x] (4) If R is a ﬁeld, or more generally a division ring, then rx = 0M implies that either r = 0R or x = 0M . [If r = 0, multiply the equation rx = 0M by r−1 .] 4.1.3 Examples 1. If M is a vector space over the ﬁeld R, then M is an R-module. 2. Any ring R is a module over itself. Rather than check all the formal requirements, think intuitively: Elements of a ring can be added and subtracted, and we can certainly multiply r ∈ R by x ∈ R, and the usual rules of arithmetic apply. 3. If R is any ring, then Rn , the set of all n-tuples with components in R, is an R-module, with the usual deﬁnitions of addition and scalar multiplication (as in Euclidean space, e.g., r(x1 , . . . , xn ) = (rx1 , . . . , rxn ), etc). 4. Let M = Mmn (R) be the set of all m × n matrices with entries in R. Then M is an R-module, where addition is ordinary matrix addition, and multiplication of the scalar c by the matrix A means multiplication of each entry of A by c. 5. Every abelian group A is a Z-module. Addition and subtraction is carried out according to the group structure of A; the key point is that we can multiply x ∈ A by the integer n. If n > 0, then nx = x + x + · · · + x (n times); if n < 0, then nx = −x − x − · · · − x (|n| times).

page 2 of Chapter 4

In all of these examples, we can switch from left to right modules by a simple notational change. This is deﬁnitely not the case in the next example. 6. Let I be a left ideal of the ring R; then I is a left R-module. (If x ∈ I and r ∈ R then rx (but not necessarily xr) belongs to I.) Similarly, a right ideal is a right R-module, and a two-sided ideal is both a left and a right R-module. An R-module M permits addition of vectors and scalar multiplication. If multiplication of vectors is allowed, we have an R-algebra. 4.1.4 Deﬁnitions and Comments Let R be a commutative ring. We say that M is an algebra over R, or that M is an R-algebra, if M is an R-module that is also a ring (not necessarily commutative), and the ring and module operations are compatible, i.e., r(xy) = (rx)y = x(ry) for all x, y ∈ M and r ∈ R. 4.1.5 Examples 1. Every commutative ring R is an algebra over itself (see Example 2 of (4.1.3)). 2. An arbitrary ring R is always a Z-algebra (see Example 5 of (4.1.3)). 3. If R is a commutative ring, then Mn (R), the set of all n × n matrices with entries in R, is an R-algebra (see Example 4 of (4.1.3)). 4. If R is a commutative ring, then the polynomial ring R[X] is an R-algebra, as is the ring R[[X]] of formal power series; see Examples 5 and 6 of (2.1.3). The compatibility condition is satisﬁed because an element of R can be regarded as a polynomial of degree 0. 5. If E/F is a ﬁeld extension, then E is an algebra over F . This continues to hold if E is a division ring, and in this case we say that E is a division algebra over F . To check that a subset S of a vector space is a subspace, we verify that S is closed under addition of vectors and multiplication of a vector by a scalar. Exactly the same idea applies to modules and algebras. 4.1.6 Deﬁnitions and Comments If N is a nonempty subset of the R-module M , we say that N is a submodule of M (notation N ≤ M ) if for every x, y ∈ N and r, s ∈ R, we have rx + sy ∈ N . If M is an R-algebra, we say that N is a subalgebra if N is a submodule that is also a subring. For example, if A is an abelian group (= Z-module), the submodules of A are the subsets closed under addition and multiplication by an integer (which amounts to addition also). Thus the submodules of A are simply the subgroups. If R is a ring, hence a module over itself, the submodules are those subsets closed under addition and also under multiplication by any r ∈ R, in other words, the left ideals. (If we take R to be a right R-module, then the submodules are the right ideals.) We can produce many examples of subspaces of vector spaces by considering kernels and images of linear transformations. A similar idea applies to modules. 4.1.7 Deﬁnitions and Comments Let M and N be R-modules. A module homomorphism (also called an R-homomorphism) from M to N is a map f : M → N such that f (rx + sy) = rf (x) + sf (y) for all x, y ∈ M and r, s ∈ R. Equivalently, f (x + y) = f (x) + f (y) and f (rx) = rf (x) for all x, y ∈ M and r ∈ R. The kernel of a homomorphism f is ker f = {x ∈ M : f (x) = 0}, and the image of f is {f (x) : x ∈ M }.

page 3 of Chapter 4

If follows from the deﬁnitions that the kernel of f is a submodule of M , and the image of f is a submodule of N . If M and N are R-algebras, an algebra homomorphism or homomorphism of algebras from M to N is an R-module homomorphism that is also a ring homomorphism. 4.1.8 Another Way to Describe an Algebra Assume that A is an algebra over the commutative ring R, and consider the map r → r1 of R into A. The commutativity of R and the compatibility of the ring and module operations imply that the map is a ring homomorphism. To see this, note that if r, s ∈ R then (rs)1 = (sr)1 = s(r1) = s[(r1)1] = (r1)(s1). Furthermore, if y ∈ A then (r1)y = r(1y) = r(y1) = y(r1) so that r1 belongs to the center of A, i.e., the set of elements that commute with everything in A. Conversely, if f is a ring homomorphism from the commutative ring R to the center of the ring A, we can make A into an R-module via rx = f (r)x. The compatibility conditions are satisﬁed because r(xy) = f (r)(xy) = (f (r)x)y = (rx)y and (f (r)x)y = (xf (r))y = x(f (r)y) = x(ry). Because of this result, the deﬁnition of an R-algebra is sometimes given as follows. The ring A is an algebra over the commutative ring R if there exists a ring homomorphism of R into the center of A. For us at this stage, such a deﬁnition would be a severe overdose of abstraction. Notational Convention: We will often write the module {0} (and the ideal {0} in a ring) simply as 0. Problems For Section 4.1 1. If I is an ideal of the ring R, show how to make the quotient ring R/I into a left Rmodule, and also show how to make R/I into a right R-module. 2. Let A be a commutative ring and F a ﬁeld. Show that A is an algebra over F if and only if A contains (an isomorphic copy of) F as a subring. Problems 3,4 and 5 illustrate that familiar properties of vector spaces need not hold for modules. 3. Give an example of an R-module M with nonzero elements r ∈ R and x ∈ M such that rx=0. 4. Let M be the additive group of rational numbers. Show that any two elements of M are linearly dependent (over the integers Z). 5. Continuing Problem 4, show that M cannot have a basis, that is, a linearly independent spanning set over Z. 6. Prove the modular law for subgroups of a given group G: With the group operation written multiplicatively, A(B ∩ C) = (AB) ∩ C if A ⊆ C. Switching to additive notation, we have, for submodules of a given R-module, A + (B ∩ C) = (A + B) ∩ C,

page 4 of Chapter 4

again if A ⊆ C. 7. Let T be a linear transformation on the vector space V over the ﬁeld F . Show how to make V into an R-module in a natural way, where R is the polynomial ring F [X]. 4.2 The Isomorphism Theorems For Modules If N is a submodule of the R-module M (notation N ≤ M ), then in particular N is an additive subgroup of M , and we may form the quotient group M/N in the usual way. In fact M/N becomes an R-module if we deﬁne r(x + N ) = rx + N . (This makes sense because if x belongs to the submodule N , so does rx.) Since scalar multiplication in the quotient module M/N is carried out via scalar multiplication in the original module M , we can check the module axioms without diﬃculty. The canonical map π : M → M/N is a module homomorphism with kernel N . Just as with groups and rings, we can establish the basic isomorphism theorems for modules. 4.2.1 Factor Theorem For Modules Any module homomorphism f : M → M whose kernel contains N can be factored through M/N . In other words, there is a unique module homomorphism f : M/N → M such that f (x + N ) = f (x). Furthermore, (i) f is an epimorphism if and only if f is an epimorphism; (ii) f is a monomorphism if and only if ker f = N ; (iii) f is an isomorphism if and only if f is a epimorphism and kerf = N . Proof. Exactly as in (2.3.1), with appropriate notational changes. (In Figure 2.3.1, replace R by M , S by M and I by N .) ♣ 4.2.2 First Isomorphism Theorem For Modules If f : M → M is a module homomorphism with kernel N , then the image of f is isomorphic to M/N . Proof. Apply the factor theorem, and note that f is an epimorphism onto its image. ♣ 4.2.3 Second Isomorphism Theorem For Modules Let S and T be submodules of M , and let S + T = {x + y : x ∈ S, y ∈ T }. Then S + T and S ∩ T are submodules of M and (S + T )/T ∼ = S/(S ∩ T ). Proof. The module axioms for S + T and S ∩ T can be checked in routine fashion. Deﬁne a map f : S → M/T by f (x) = x + T . Then f is a module homomorphism whose kernel is S ∩ T and whose image is {x + T : x ∈ S} = (S + T )/T . The ﬁrst isomorphism theorem for modules gives the desired result. ♣ 4.2.4 Third Isomorphism Theorem For Modules If N ≤ L ≤ M , then M/L ∼ = (M/N )/(L/N ). Proof. Deﬁne f : M/N → M/L by f (x + N ) = x + L. As in (2.3.4), the kernel of f is {x + N : x ∈ L} = L/N , and the image of f is {x + L : x ∈ M } = M/L. The result follows from the ﬁrst isomorphism theorem for modules. ♣ 4.2.5 Correspondence Theorem For Modules Let N be a submodule of the R-module M . The map S → S/N sets up a one-to-one correspondence between the set of all submodules of M containing N and the set of all submodules of M/N . The inverse of the map is T → π −1 (T ), where π is the canonical map: M → M/N . Proof. The correspondence theorem for groups yields a one-to-one correspondence between additive subgroups of M containing N and additive subgroups of M/N . We must check that submodules correspond to submodules, and it is suﬃcient to show that if S1 /N ≤ S2 /N , then S1 ≤ S2 (the converse is immediate). If x ∈ S1 , then x + N ∈ S1 /N ⊆ S2 /N , so x + N = y + N for some y ∈ S2 . Thus x − y ∈ N ⊆ S2 , and since y ∈ S2 we must have x ∈ S2 as well. Therefore S1 ≤ S2 . ♣

page 5 of Chapter 4

We now look at modules that have a particularly simple structure, and can be used as building blocks for more complicated modules. 4.2.6 Deﬁnitions and Comments An R-module M is cyclic if it is generated by a single element x. In other words, M = Rx = {rx : r ∈ R}. Thus every element of M is a scalar multiple of x. (If x = 0, then M = {0}, which is called the zero module and is often written simply as 0.) A cyclic vector space over a ﬁeld is a one-dimensional space, assuming that x = 0. The annihilator of an element y in the R-module M is Iy = {r ∈ R : ry = 0}, a left ideal of R. If M is cyclic with generator x, then M ∼ = R/Ix . To see this, apply the ﬁrst isomorphism theorem for modules to the map r → rx of R onto M . The annihilator of the module M is Io = {r ∈ R : ry = 0 for every y ∈ M }. Note that Io is a two-sided ideal, because if r ∈ I0 and s ∈ R, then for every y ∈ M we have (rs)y = r(sy) = 0. When R is commutative, annihilating the generator of a cyclic module is equivalent to annihilating the entire module. 4.2.7 Lemma (a) If x generates a cyclic module M over the commutative ring R, then Ix = Io , so that M∼ = R/Io . (In this situation, Io is frequently referred to as the order ideal of M .) (b) Two cyclic R-modules over a commutative ring are isomorphic if and only if they have the same annihilator. Proof. (a) If rx = 0 and y ∈ M , then y = sx for some x ∈ R, so ry = r(sx) = s(rx) = s0 = 0. (b) The “if” part follows from (a), so assume that g : Rx → Ry is an isomorphism of cyclic R-modules. Since g is an isomorphism, g(x) must be a generator of Ry, so we may as well take g(x) = y. Then g(rx) = rg(x) = ry, so rx corresponds to ry under the isomorphism. Therefore r belongs to the annihilator of Rx if and only if r belongs to the annihilator of Ry. ♣ Problems For Section 4.2 1. Show that every submodule of the quotient module M/N can be expressed as (L + N )/N for some submodule L of M . 2. In Problem 1, must L contain N ? 3. In the matrix ring Mn (R), let M be the submodule generated by E11 , the matrix with 1 in row 1, column 1, and 0’s elsewhere. Thus M = {AE11 : A ∈ Mn (R)}. Show that M consists of all matrices whose entries are zero except perhaps in column 1. 4. Continuing Problem 3, show that the annihilator of E11 consists of all matrices whose ﬁrst column is zero, but the annihilator of M is {0}. 5. If I is an ideal of the ring R, show that the quotient ring is a cyclic R-module. 6. Let M be an R-module, and let I be an ideal of R. We wish to make M into an R/Imodule via (r + I)m = rm, r ∈ R, m ∈ M . When will this be legal? 7. Assuming legality in Problem 6, let M1 be the resulting R/I-module, and note that as sets, M1 = M . Let N be a subset of M and consider the following two statements: (a) N is an R-submodule of M ; (b) N is an R/I-submodule of M1 . Can one of these statements be true and the other false? 4.3 Direct Sums and Free Modules 4.3.1 Direct Products In Section 1.5, we studied direct products of groups, and the basic idea seems to carry over to modules. Suppose that we have an R-module Mi for each i in some index set I (possibly inﬁnite). The members of the direct product of the Mi , denoted by i∈I Mi , are all families (ai , i ∈ I). (A family is just a function on I whose value at the

page 6 of Chapter 4

element i is ai .) Addition is described by (ai ) + (bi ) = (ai + bi ) and scalar multiplication by r(ai ) = (rai ). There is nothing wrong with this deﬁnition, but the resulting mathematical object has some properties that are undesirable. If (ei ) is the family with 1 in position i and zeros elsewhere, then (thinking about vector spaces) it would be useful to express an arbitrary element of the direct product in terms of the ei . But if the index set I is inﬁnite, we will need concepts of limit and convergence, and this will take us out of algebra and into analysis. Another approach is to modify the deﬁnition of direct product. 4.3.2 Deﬁnitions The external direct sum of the modules Mi , i ∈ I, denoted by ⊕i∈I Mi , consists of all families (ai , i ∈ I) such that ai = 0 for all but ﬁnitely many i. Addition and scalar multiplication are deﬁned exactly as for the direct product, so that the external direct sum coincides with the direct product when the index set I is ﬁnite. The R-module M is the internal direct sum of the submodules Mi if each x ∈ M can be expressed uniquely as xi1 + · · · + xin where 0 = xik ∈ Mik , k = 1, . . . , n. (The positive integer n and the elements xik depend on x. In any expression of this type, the indices ik are assumed distinct.) Just as with groups, the internal and external direct sums are isomorphic. To see this without a lot of formalism, let the element xik ∈ Mik correspond to the family that has xik in position ik and zeros elsewhere. We will follow standard practice and refer to the “direct sum” without the qualifying adjective. Again as with groups, the next result may help us to recognize when a module can be expressed as a direct sum. 4.3.3 Proposition The module M is the direct sum of submodules Mi if and only if both of the following conditions are satisﬁed: (1) M = i Mi , that is, each x ∈ M is a ﬁnite sum of the form xi1 + · · · + xin , where xik ∈ Mik ; (2) For each i, Mi ∩ j=i Mj = 0. (Note that in condition (1), we do not assume that the representation is unique. Observe also that another way of expressing (2) is that if xi1 + · · · + xin = 0, with xik ∈ Mik , then xik = 0 for all k.) Proof. The necessity of the conditions follows from the deﬁnition of external direct sum, so assume that (1) and (2) hold. If x ∈ M then by (1), x is a ﬁnite sum of elements from various Mi ’s. For convenience in notation, say x = x1 + x2 + x3 + x4 with xi ∈ Mi , i = 1, 2, 3, 4. If the representation is not unique, say x = y1 + y2 + y4 + y5 + y6 with yi ∈ Mi , i = 1, 2, 4, 5, 6. Then x3 is a sum of terms from modules other than M3 , so by (2), x3 = 0. Similarly, y5 = y6 = 0 and we have x1 + x2 + x4 = y1 + y2 + y4 . But then x1 − y1 is a sum of terms from modules other than M1 , so by (2), x1 = y1 . Similarly x2 = y2 , x4 = y4 , and the result follows. ♣ A basic property of the direct sum M = ⊕i∈I Mi is that homomorphisms fi : Mi → N can be “lifted” to M . In other words, there is a unique homomorphism f : M → N such that for each i, f = fi on Mi . Explicitly, f (xi1 + · · · + xir ) = fi1 (xi1 ) + · · · + fir (xir ). [No other choice is possible for f , and since each fi is a homomorphism, so is f .] We know that every vector space has a basis, but not every module is so fortunate; see Section 4.1, Problem 5. We now examine modules that have this feature. 4.3.4 Deﬁnitions and Comments Let S be a subset of the R-module M . We say that S is linearly independent over R if λ1 x1 + · · · + λk xk = 0 implies that all λi = 0 (λi ∈ R, xi ∈

page 7 of Chapter 4

S, k = 1, 2, . . .). We say that S is a spanning (or generating) set for M over R, or that S spans (generates) M over R if each x ∈ M can be written as a ﬁnite linear combination of elements of S with coeﬃcients in R. We will usually omit “over R” if the underlying ring R is clearly identiﬁed. A basis is a linearly independent spanning set, and a module that has a basis is said to be free. Suppose that M is a free module with basis (bi , i ∈ I), and we look at the submodule Mi spanned by the basis element bi . (In general, the submodule spanned (or generated) by a subset T of M consists of all ﬁnite linear combinations of elements of T with coeﬃcients in R. Thus the submodule spanned by bi is the set of all rbi , r ∈ R.) If R is regarded as a module over itself, then the map r → rbi is an R-module isomorphism of R and Mi , because {bi } is a linearly independent set. Since the bi span M , it follows that M is the sum of the submodules Mi , and by linear independence of the bi , the sum is direct. Thus we have an illuminating interpretation of a free module: A free module is a direct sum of isomorphic copies of the underlying ring R. Conversely, a direct sum of copies of R is a free R-module. If ei has 1 as its ith component and zeros elsewhere, the ei form a basis. This characterization allows us to recognize several examples of free modules. 1. For any positive integer n, Rn is a free R-module. 2. The matrix ring Mmn (R) is a free R-module with basis Eij , i = 1, . . . , m, j = 1, . . . , n. 3. The polynomial ring R[X] is a free R-module with basis 1, X, X 2 , . . .. We will adopt the standard convention that the zero module is free with the empty set as basis. Any two bases for a vector space over a ﬁeld have the same cardinality. This property does not hold for arbitrary free modules, but the following result covers quite a few cases. 4.3.5 Theorem Any two bases for a free module M over a commutative ring R have the same cardinality. Proof. If I is a maximal ideal of R, then k = R/I is a ﬁeld, and V = M/IM is a vector space over k. [By IM we mean all ﬁnite sums ai xi with ai ∈ I and xi ∈ M ; thus IM is a submodule of M . If r + I ∈ k and x + IM ∈ M/IM , we take (r + I)(x + IM ) to be rx + IM . The scalar multiplication is well-deﬁned because r ∈ I or x ∈ IM implies that rx ∈ IM . We can express this in a slightly diﬀerent manner by saying that I annihilates M/IM . The requirements for a vector space can be checked routinely.] . Since the xi span M , the xi span M/IM . Now if (xi ) is a basis for M , let xi = xi +IM ai xi = 0, where ai = ai + I, ai ∈ R, then ai xi ∈ IM . Thus ai xi = bj xj with If bj ∈ I. Since the xi form a basis, we must have ai = bj for some j. Consequently ai ∈ I, so that ai = 0 in k. We conclude that the xi form a basis for V over k, and since the dimension of V over k depends only on M , R and I, and not on a particular basis for M , the result follows. ♣ 4.3.6 Some Key Properties of Free Modules Suppose that M is a free module with basis (xi ), and we wish to construct a module homomorphism f from M to an arbitrary module N . Just as with vector spaces, we can specify f (xi ) = yi ∈ N arbitrarily on basis elements, and extend by linearity. Thus if x = ai xi ∈ M , we have f (x) = ai yi . (The idea should be familiar; for example, a linear transformation on Euclidean 3-space is determined by what it does to the three standard basis vectors.) Now let’s turn this process around: If N is an arbitrary module, we can express N as a homomorphic image of a free module. All we need is a set (yi , i ∈ I) of generators for N . (If all else fails, we can take the yi to be all the elements of N .) We then construct a free module with basis (xi , i ∈ I). (To do this,

page 8 of Chapter 4

take the direct sum of copies of R, as many copies as there are elements of I.) Then map xi to yi for each i. Note that by the ﬁrst isomorphism theorem, every module is a quotient of a free module. Problems For Section 4.3 1. Showthat in Proposition 4.3.3, (2) can be replaced by the weaker condition that for each i, Mi ∩ j
m

aij wi , j = 1, . . . , n

(1)

i=1

where the aij belong to R. It is natural to associate the m × n matrix A with the homomorphism f , and it appears that we have an isomorphism of some sort, but an isomorphism of what? If f and g are homomorphisms of M into N , then f and g can be added (and subtracted): (f + g)(x) = f (x) + g(x). If f is represented by the matrix A and g by B, then f + g corresponds to A+B. This gives us an abelian group isomorphism of HomR (M, N ), the set of all R-module homomorphisms from M to N , and Mmn (R), the set of all m × n matrices with entries in R. In addition, Mmn (R) is an R-module, so it is tempting to say “obviously, we have an R-module isomorphism”. But we must be very careful here. If f ∈ HomR (M, N ) and s ∈ R, we can deﬁne sf in the natural way: (sf )(x) = sf (x). However, if we carry out the “routine” check that sf ∈ HomR (M, N ), there is one step that causes alarm bells to go oﬀ: (sf )(rx) = sf (rx) = srf (x), but r(sf )(x) = rsf (x) and the two expressions can disagree if R is not commutative. Thus HomR (M, N ) need not be an R-module. Let us summarize what we have so far.

page 9 of Chapter 4

4.4.1 The Correspondence Between Homomorphisms and Matrices Associate with each f ∈ HomR (M, N ) a matrix A as in (1) above. This yields an abelian group isomorphism, and also an R-module isomorphism if R is commutative. Now let m = n, so that the dimensions are equal and the matrices are square, and take vi = wi for all i. A homomorphism from M to itself is called an endomorphism of M , and we use the notation EndR (M ) for HomR (M, M ). Since EndR (M ) is a ring under composition of functions, and Mn (R) is a ring under matrix multiplication, it is plausible to conjecture that we have a ring isomorphism. If f corresponds to A and g to B, then we can apply g to both sides of (1) to obtain g(f (vj )) =

n i=1

aij

n

bki vk =

k=1

n n ( aij bki )vk .

(2)

k=1 i=1

If R is commutative, then aij bki = bki aij , and the matrix corresponding to gf = g ◦ f is BA, as we had hoped. In the noncommutative case, we will not be left empty-handed if we deﬁne the opposite ring Ro , which has exactly the same elements as R and the same addition structure. However, multiplication is done backwards, i.e., ab in Ro is ba in R. It is convenient to attach a superscript o to the elements of Ro , so that ao bo = ba (more precisely, ao bo = (ba)o ). Thus in (2) we have aij bki = boki aoij . To summarize, The endomorphism ring EndR (M ) is isomorphic to the ring of n × n matrices with coeﬃcients in the opposite ring Ro . If R is commutative, then EndR (M ) is ring-isomorphic to Mn (R). 4.4.2 Preparation For The Smith Normal Form We now set up some basic machinery to be used in connection with the Smith normal form and its applications. Assume that M is a free Z-module of rank n, with basis x1 , . . . , xn , and that K is a submodule of M with ﬁnitely many generators u1 , . . . , um . (We say that K is ﬁnitely generated.) We change to a new basis y1 , . . . , yn via Y = P X, where X [resp. Y ] is a column vector with components xi [resp.yi ]. Since X and Y are bases, the n × n matrix P must be invertible, and we need to be very clear on what this means. If the determinant of P is nonzero, we can construct P −1 , for example by the “adjoint divided by determinant” formula given in Cramer’s rule. But the underlying ring is Z, not Q, so we require that the coeﬃcients of P −1 be integers. (For a more transparent equivalent condition, see Problem 1.) Similarly, we are going to change generators of K via V = QU , where Q is an invertible m × m matrix and U is a column vector with components ui . The generators of K are linear combinations of basis elements, so we have an equation of the form U = AX, where A is an m × n matrix called the relations matrix. Thus V = QU = QAX = QAP −1 Y. so the new relations matrix is

B = QAP −1 .

Thus B is obtained from A by pre-and postmultiplying by invertible matrices, and we say that A and B are equivalent. We will see that two matrices are equivalent iﬀ they have the same Smith normal form. The point we wish to emphasize now is that if we know the matrix P , we can compute the new basis Y , and if we know the matrix Q, we can compute the new system of generators V . In our applications, P and Q will be constructed by elementary row and column operations.

page 10 of Chapter 4

Problems For Section 4.4 1. Show that a square matrix P over the integers has an inverse with integer entries if and only if P is unimodular, that is, the determinant of P is ±1. 2. Let V be the direct sum of the R-modules V1 , . . . , Vn , and let W be the direct sum of R-modules W1 , . . . , Wm . Indicate how a module homomorphism from V to W can be represented by a matrix. (The entries of the matrix need not be elements of R.) 3. Continuing Problem 2, show that if V n is the direct sum of n copies of the R-module V , then we have a ring isomorphism EndR (V n ) ∼ = Mn (EndR (V )). 4. Show that if R is regarded as an R-module, then EndR (R) is isomorphic to the opposite ring Ro . 5. Let R be a ring, and let f ∈ EndR (R). Show that for some r ∈ R we have f (x) = xr for all x ∈ R. 6. Let M be a free R-module of rank n. Show that EndR (M ) ∼ = Mn (Ro ), a ring isomorphism. 7. Continuing Problem 6, if R is commutative, show that the ring isomorphism is in fact an R-algebra isomorphism. 4.5 Smith Normal Form We are going to describe a procedure that is very similar to reduction of a matrix to echelon form. The result is that every matrix over a principal ideal domain is equivalent to a matrix in Smith normal form. Explicitly, the Smith matrix has nonzero entries only on the main diagonal. The main diagonal entries are, from the top, a1 , . . . , ar (possibly followed by zeros), where the ai are nonzero and ai divides ai+1 for all i. We will try to convey the basic ideas via a numerical example. This will allow us to give informal but convincing proofs of some major theorems. A formal development is given in Jacobson, Basic Algebra I, Chapter 3. All our computations will be in the ring of integers, but we will indicate how the results can be extended to an arbitrary principal ideal domain. Let’s start with the following matrix:   0 0 22 0  −2 2 −6 −4  2 2 6 8 As in (4.4.2), we assume a free Z-module M with basis x1 , x2 , x3 , x4 , and a submodule K generated by u1 , u2 , u3 , where u1 = 22x3 , u2 = −2x1 + 2x2 − 6x3 − 4x4 , u3 = 2x1 + 2x2 + 6x3 + 8x4 . The ﬁrst step is to bring the smallest positive integer to the 1-1 position. Thus interchange rows 1 and 3 to obtain   2 2 6 8  −2 2 −6 −4  0 0 22 0 Since all entries in column 1, and similarly in row 1, are divisible by 2, we can pivot about the 1-1 position, in other words, use the 1-1 entry to produce zeros. Thus add row 1 to row 2 to get   2 2 6 8 0 4 0 4 0 0 22 0 Add -1 times column 1 to column 2, then add -3 times column 1 to column 4. The result is  2 0 0 0 4 0 0 0 22

times column 1 to column 3, and add -4  0 4 0

page 11 of Chapter 4

Now we have “peeled oﬀ” the ﬁrst row and column, and we bring the smallest positive integer to the 2-2 position. It’s already there, so no action is required. Furthermore, the 2-2 element is a multiple of the 1-1 element, so again no action is required. Pivoting about the 2-2 position, we add -1 times column 2 to column 4, and we have   2 0 0 0 0 4 0 0 0 0 22 0 Now we have peeled oﬀ the ﬁrst two rows and columns, and we bring the smallest positive integer to the 3-3 position; again it’s already there. But 22 is not a multiple of 4, so we have more work to do. Add row 3 to row 2 to get   2 0 0 0  0 4 22 0  0 0 22 0 Again we pivot about the 2-2 position; 4 2 to column 3, we have  2 0 0

does not divide 22, but if we add -5 times column 0 4 0

0 2 22

 0 0 0

Interchange columns 2 and 3 to get 

2 0 0

0 2 22

Add -11 times row 2 to row 3 to obtain  2 0 0 2 0 0

 0 0 4 0 0 0

0 4 −44

 0 0 0

Finally, add -2 times column 2 to column 3, and then (as a convenience to get rid of the minus sign) multiply row (or column) 3 by -1; the result is   2 0 0 0 0 2 0 0 0 0 44 0 which is the Smith normal form of the original matrix. Although we had to backtrack to produce a new pivot element in the 2-2 position, the new element is smaller than the old one (since it is a remainder after division by the original number). Thus we cannot go into an inﬁnite loop, and the algorithm will indeed terminate in a ﬁnite number of steps. In view of (4.4.2), we have the following interpretation. We have a new basis y1 , y2 , y3 , y4 for M , and new generators v1 , v2 , v3 for K, where v1 = 2y1 , v2 = 2y2 , and v3 = 44y3 . In fact since the vj ’s are nonzero multiples of the corresponding yj ’s, they are linearly independent, and consequently form a basis of K. The new basis and set of generators can be expressed in terms of the original sets; see Problems 1-3 for the technique. The above discussion indicates that the Euclidean algorithm guarantees that the Smith normal form can be computed in ﬁnitely many steps. Therefore the Smith procedure can be carried out in any Euclidean domain. In fact we can generalize to a principal ideal

page 12 of Chapter 4

domain. Suppose that at a particular stage of the computation, the element a occupies the 1-1 position of the Smith matrix S, and the element b is in row 1, column 2. To use a as a pivot to eliminate b, let d be the greatest common divisor of a and b, and let r and s be elements of R such that ar + bs = d (see (2.7.2)). We postmultiply the Smith matrix by a matrix T of the following form (to aid in the visualization, we give a concrete 5 × 5 example):   r b/d 0 0 0  s −a/d 0 0 0    0 1 0 0 0   0 0 0 1 0 0 0 0 0 1 The 2×2 matrix in the upper left hand corner has determinant -1, and is therefore invertible over R. The element in the 1-1 position of ST is ar + bs = d, and the element in the 1-2 position is ab/d − ba/d = 0, as desired. We have replaced the pivot element a by a divisor d, and this will decrease the number of prime factors, guaranteeing the ﬁnite termination of the algorithm. Similarly, if b were in the 2-1 position, we would premultiply S by the transpose of T ; thus in the upper left hand corner we would have

r b/d

s −a/d

Problems For Section 4.5 1. Let A be the matrix



−2  −3 −12

3 3 12

 0 0 6

over the integers. Find the Smith normal form of A. (It is convenient to begin by adding column 2 to column 1.) 2. Continuing Problem 1, ﬁnd the matrices P and Q, and verify that QAP −1 is the Smith normal form. 3. Continuing Problem 2, if the original basis for M is {x1 , x2 , x3 } and the original set of generators of K is {u1 , u2 , u3 }, ﬁnd the new basis and set of generators. It is intuitively reasonable, but a bit messy to prove, that if a matrix A over a PID is multiplied by an invertible matrix, then the greatest common divisor of all the i × i minors of A is unchanged. Accept this fact in doing Problems 4 and 5. 4. The nonzero components ai of the Smith normal form S of A are called the invariant factors of A. Show that the invariant factors of A are unique. 5. Show that two m × n matrices are equivalent if and only if they have the same invariant factors, i.e. (by Problem 4), if and only if they have the same Smith normal form. 6. Recall that when a matrix over a ﬁeld is reduced to row-echelon form (only row operations are involved), a pivot column is followed by non-pivot columns whose entries are zero in all rows below the pivot element. When a similar computation is carried out over the integers, or more generally over a Euclidean domain, the resulting matrix is said to be in Hermite normal form. We indicate the procedure in a typical example. Let 

6 A= 9 12 Carry out the following sequence of steps: 1. Add -1 times row 1 to row 2

4 6 8

13 0 −1

 5 7 . 12

page 13 of Chapter 4

2. Interchange rows 1 and 2 3. Add -2 times row 1 to row 2, and then add -4 times row 1 to row 3 4. Add -1 times row 2 to row 5. Interchange rows 2 and 3 6. Add -3 times row 2 to row 3 7. Interchange rows 2 and 3 8. Add -4 times row 2 to row 3 9. Add 5 times row 2 to row 1 (this corresponds to choosing 0, 1, ..., m − 1 as a complete system of residues mod m) 10. Add 2 times row 3 to row 1, and then add row 3 to row 2 We now have reduced A to Hermite normal form. 7. Continuing Problem 6, consider the simultaneous equations 6x + 4y + 13z ≡ 5, 9x + 6y ≡ 7, 12x + 8y − z ≡ 12 (mod m) For which values of m ≥ 2 will the equations be consistent? 4.6 Fundamental Structure Theorems The Smith normal form yields a wealth of information about modules over a principal ideal domain. In particular, we will be able to see exactly what ﬁnitely generated abelian groups must look like. Before we proceed, we must mention a result that we will use now but not prove until later (see (7.5.5), Example 1, and (7.5.9)). If M is a ﬁnitely generated module over a PID R, then every submodule of M is ﬁnitely generated. [R is a Noetherian ring, hence M is a Noetherian R-module.] To avoid gaps in the current presentation, we can restrict our attention to ﬁnitely generated submodules. 4.6.1 Simultaneous Basis Theorem Let M be a free module of ﬁnite rank n ≥ 1 over the PID R, and let K be a submodule of M . Then there is a basis {y1 , . . . , yn } for M and nonzero elements a1 , . . . , ar ∈ R such that r ≤ n, ai divides ai+1 for all i, and {a1 y1 , . . . , ar yr } is a basis for K. Proof. This is a corollary of the construction of the Smith normal form, as explained in Section 4.5. ♣ 4.6.2 Corollary Let M be a free module of ﬁnite rank n over the PID R. Then every submodule of M is free of rank at most n. Proof. By (4.6.1), the submodule K has a basis with r ≤ n elements. ♣ In (4.6.2), the hypothesis that M has ﬁnite rank can be dropped, as the following sketch suggests. We can well-order the generators uα of K, and assume as a transﬁnite induction hypothesis that for all β < α, the submodule Kβ spanned by all the generators up to uβ is free of rank at most that of M , and that if γ < β, then the basis of Kγ is contained in the basis of Kβ . The union of the bases Sβ of the Kβ is a basis Sα for Kα . Furthermore, the inductive step preserves the bound on rank. This is because |Sβ | ≤ rank M for all β < α, and |Sα | is the smallest cardinal bounded below by all |Sβ |, β < α. Thus |Sα | ≤ rank M . 4.6.3 Fundamental Decomposition Theorem Let M be a ﬁnitely generated module over the PID R. Then there are ideals I1 =< a1 >, I2 =< a2 >, . . . , In =< an > of R such that I1 ⊇ I2 ⊇ . . . ⊇ In (equivalently, a1 |a2 | . . . |an ) and M∼ = R/I1 ⊕ R/I2 ⊕ · · · ⊕ R/In . Thus M is a direct sum of cyclic modules.

page 14 of Chapter 4

Proof. By (4.3.6), M is the image of a free module Rn under a homomorphism f . If K is the kernel of f , then by (4.6.1) we have a basis y1 , . . . , yn for Rn and a corresponding basis a1 y1 , . . . , ar yr for K. We set ai = 0 for r < i ≤ n. Then M∼ = Rn /K ∼ =

Ry1 ⊕ · · · ⊕ Ryn ∼ Ryi /Rai yi . = Ra1 y1 ⊕ · · · ⊕ Ran yn i=1 n

(To justify the last step, apply the ﬁrst isomorphism theorem to the map r1 y1 + · · · + rn yn → (r1 y1 + Ra1 y1 , . . . , rn yn + Ran yn .) But

Ryi /Rai yi ∼ = R/Rai ,

as can be seen via an application of the ﬁrst isomorphism theorem to the map r → ryi + Rai yi . Thus if Ii = Rai , i = 1, . . . , n, we have M∼ =

n

R/Ii

i=1

and the result follows. ♣ Remark It is plausible, and can be proved formally, that the uniqueness of invariant factors in the Smith normal form implies the uniqueness of the decomposition (4.6.3). Intuitively, the decomposition is completely speciﬁed by the sequence a1 , . . . , an , as the proof of (4.6.3) indicates. 4.6.4 Finite Abelian Groups Suppose that G is a ﬁnite abelian group of order 1350; what can we say about G? In the decomposition theorem (4.6.3), the components of G are of the form Z/Zai , that is, cyclic groups of order ai . We must have ai |ai+1 for all i, and since the order of a direct sum is the product of the orders of the components, we have a1 · · · ar = 1350. The ﬁrst step in the analysis is to ﬁnd the prime factorization of 1350, which is (2)(33 )(52 ). One possible choice of the ai is a1 = 3, a2 = 3, a3 = 150. It is convenient to display the prime factors of the ai , which are called elementary divisors, as follows: a1 = 3 = 20 31 50 a2 = 3 = 20 31 50 a3 = 150 = 21 31 52 Since a1 a2 a3 = 21 33 52 , the sum of the exponents of 2 must be 1, the sum of the exponents of 3 must be 3, and the sum of the exponents of 5 must be 2. A particular distribution of exponents of a prime p corresponds to a partition of the sum of the exponents. For example, if the exponents of p were 0,1,1 and 2, this would correspond to a partition of 4 as 1+1+2. In the above example, the partitions are 1=1, 3=1+1+1, 2=2. We can count the number of abelian groups of order 1350 (up to isomorphism) by counting partitions. There is only one partition of 1, there are two partitions of 2 (2 and 1+1) and three partitions of 3 (3,1+2 and 1+1+1). [This pattern does not continue; there are ﬁve partitions of 4, namely 4,1+3,1+1+2,1+1+1+1,2+2, and seven partitions of 5, namely 5,1+4,1+1+3,1+1+1+2,1+1+1+1+1,1+2+2,2+3.] We specify a group by choosing a partition of 1, a partition of 3 and a partition of 2, and the number of possible choices is (1)(3)(2)=6. Each choice of a sequence of partitions produces a diﬀerent sequence of invariant factors. Here is the entire list; the above example appears as entry (5). (1) a1 = 21 33 52 = 1350, G ∼ = Z1350

page 15 of Chapter 4

(2) a1 = 20 30 51 = 5, a2 = 21 33 51 = 270, G ∼ = Z5 ⊕ Z270 (3) a1 = 20 31 50 = 3, a2 = 21 32 52 = 450, G ∼ = Z3 ⊕ Z450 (4) a1 = 20 31 51 = 15, a2 = 21 32 51 = 90, G ∼ = Z15 ⊕ Z90 (5) a1 = 20 31 50 = 3, a2 = 20 31 50 = 3, a3 = 21 31 52 = 150, G ∼ = Z3 ⊕ Z3 ⊕ Z150 (6) a1 = 20 31 50 = 3, a2 = 20 31 51 = 15, a3 = 21 31 51 = 30, G ∼ = Z3 ⊕ Z15 ⊕ Z30 . In entry (6) for example, the maximum number of summands in a partition is 3 (=1+1+1), and this reveals that there will be three invariant factors. The partition 2=1+1 has only two summands, and it is “pushed to the right” so that 51 appears in a2 and a3 but not a1 . (Remember that we must have a1 |a2 |a3 .). Also, we can continue to decompose some of the components in the direct sum representation of G. (If m and n are relatively prime, then Zmn ∼ = Zm ⊕ Zn by the Chinese remainder theorem.) However, this does not change the conclusion that there are only 6 mutually nonisomorphic abelian groups of order 1350. Before examining inﬁnite abelian groups, let’s come back to the fundamental decomposition theorem. 4.6.5 Deﬁnitions and Comments If x belongs to the R-module M , where R is any integral domain, then x is a torsion element if rx = 0 for some nonzero r ∈ R. The torsion submodule T of M is the set of torsion elements. (T is indeed a submodule; if rx = 0 and sy = 0, then rs(x + y) = 0.) M is a torsion module if T is all of M , and M is torsion-free if T consists of 0 alone, in other words, rx = 0 implies that either r = 0 or x = 0. A free module must be torsion-free, by deﬁnition of linear independence. Now assume that R is a PID, and decompose M as in (4.6.3), where a1 , . . . , ar are nonzero and ar+1 = · · · = an = 0. Each module R/ < ai >, 1 ≤ i ≤ r, is torsion (it is annihilated by ai ), and the R/ < ai >, r + 1 ≤ i ≤ n, are copies of R. Thus ⊕ni=r+1 R/ < ai > is free. We conclude that (*) every ﬁnitely generated module over a PID is the direct sum of its torsion submodule and a free module and (**) every ﬁnitely generated torsion-free module over a PID is free. In particular, a ﬁnitely generated abelian group is the direct sum of a number (possibly zero) of ﬁnite cyclic groups and a free abelian group (possibly {0}). 4.6.6 Abelian Groups Speciﬁed by Generators and Relations Suppose that we have a free abelian group F with basis x1 , x2 , x3 , and we impose the following constraints on the xi : (1) 2x1 + 2x2 + 8x3 = 0, −2x1 + 2x2 + 4x3 = 0. What we are doing is forming a “submodule of relations” K with generators u1 = 2x1 + 2x2 + 8x3 and u2 = −2x1 + 2x2 + 4x3

(2)

and we are identifying every element in K with zero. This process yields the abelian group G = F/K, which is generated by x1 + K, x2 + K and x3 + K. The matrix associated with (2) is 2 2 8 −2 2 4 and a brief computation gives the Smith normal form 2 0 0 . 0 4 0

page 16 of Chapter 4

Thus we have a new basis y1 , y2 , y3 for F and new generators 2y1 , 4y2 for K. The quotient group F/K is generated by y1 + K, y2 + K and y3 + K, with 2(y1 + K) = 4(y2 + K) = 0 + K. In view of (4.6.3) and (4.6.5), we must have F/K ∼ = Z2 ⊕ Z4 ⊕ Z. Canonical forms of a square matrix A can be developed by reducing the matrix xI − A to Smith normal form. In this case, R is the polynomial ring F [X] where F is a ﬁeld. But the analysis is quite lengthy, and I prefer an approach in which the Jordan canonical form is introduced at the very beginning, and then used to prove some basic results in the theory of linear operators; see Ash, A Primer of Abstract Mathematics, MAA 1998. Problems For Section 4.6 1. Classify all abelian groups of order 441. 2. Classify all abelian groups of order 40. 3. Identify the abelian group given by generators x1 , x2 , x3 and relations x1 + 5x2 + 3x3 = 0, 2x1 − x2 + 7x3 = 0, 3x1 + 4x2 + 2x3 = 0.

4. In (4.6.6), suppose we cancel a factor of 2 in Equation (1). This changes the matrix associated with (2) to 1 1 4 , −1 1 2 whose Smith normal form diﬀers from that given in the text. What’s wrong? 5. Let M , N and P be abelian groups. If M ⊕ N ∼ = M ⊕ P , show by example that N need not be isomorphic to P . 6. In Problem 5, show that M ⊕ N ∼ = M ⊕ P does imply N ∼ = P if M , N and P are ﬁnitely generated. 4.7 Exact Sequences and Diagram Chasing 4.7.1 Deﬁnitions and Comments Suppose that the R-module M is the direct sum of the submodules A and B. Let f be the inclusion or injection map of A into M (simply the identity function on A), and let g be the natural projection of M on B, given by g(a + b) = b, a ∈ A, b ∈ B. The image of f , namely A, coincides with the kernel of g, and we say that the sequence f g (1) A → M → B is exact at M . A longer (possibly inﬁnite) sequence of homomorphisms is said to be exact if it is exact at each junction, that is, everywhere except at the left and right endpoints, if they exist. There is a natural exact sequence associated with any module homomorphism g : M → N , namely f g (2) 0 → A → M → B → 0. In the diagram, A is the kernel of g, f is the injection map, and B is the image of g. A ﬁve term exact sequence with zero modules at the ends, as in (2), is called a short exact sequence. Notice that exactness at A is equivalent to ker f = 0, i.e., injectivity of f . Exactness at B is equivalent to im g = B, i.e., surjectivity of g. Notice also that by the ﬁrst isomorphism theorem, we may replace B by M/A and g by the canonical map of M onto M/A, while preserving exactness.

page 17 of Chapter 4

Now let’s come back to (1), where M is the direct sum of A and B, and attach zero modules to produce the short exact sequence (2). If we deﬁne h as the injection of B into M and e as the projection of M on A, we have (see (3) below) g ◦ h = 1 and e ◦ f = 1, where 1 stands for the identity map.

0

→

e ← A → f

M

h ← → B g

→

0

(3)

The short exact sequence (2) is said to split on the right if there is a homomorphism h : B → M such that g ◦ h = 1, and split on the left if there is a homomorphism e : M → A such that e ◦ f = 1. These conditions turn out to be equivalent, and both are equivalent to the statement that M is essentially the direct sum of A and B. “Essentially” means that not only is M isomorphic to A ⊕ B, but f can be identiﬁed with the injection of A into the direct sum, and g with the projection of the direct sum on B. We will see how to make this statement precise, but ﬁrst we must turn to diagram chasing, which is a technique for proving assertions about commutative diagrams by sliding from one vertex to another. The best way to get accustomed to the method is to do examples. We will work one out in great detail in the text, and there will be more practice in the exercises, with solutions provided. We will use the shorthand gf for g ◦ f and f m for f (m). 4.7.2 The Five Lemma Consider the following commutative diagram with exact rows. D

e →

s↓ D

A

f →

t↓ → e

A

M

g →

u↓ → f

M

B

h →

v↓ → g

B

C w↓

→ h

C

If s, t, v and w are isomorphisms, so is u. Proof. The two parts of the proof are of interest in themselves, and are frequently called the “four lemma”, since they apply to diagrams with four rather than ﬁve modules in each row. (i) If t and v are surjective and w is injective, then u is surjective. (ii) If s is surjective and t and v are injective, then u is injective. [The pattern suggests a “duality” between injective and surjective maps. This idea will be explored in Chapter 10; see (10.1.4).] The ﬁve lemma follows from (i) and (ii). To prove (i), let m ∈ M . Then g m ∈ B , and since v is surjective, we can write g m = vb for some b ∈ B. By commutativity of the square on the right, h vb = whb. But h vb = h g m = 0 by exactness of the bottom row at B , and we then have whb = 0. Thus hb ∈ ker w, and since w is injective, we have hb = 0, so that b ∈ ker h = im g by exactness of the top row at B. So we can write b = gm for some m ∈ M . Now g m = vb (see above) = vgm = g um by commutativity of the square M BB M . Therefore m − um ∈ ker g = imf by exactness of the bottom row at M . Let m − um = f a for some a ∈ A . Since t is surjective, a = ta for some a ∈ A, and by commutativity of the square AM M A , f ta = uf a, so m − um = uf a, so m = u(m + f a). Consequently, m belongs to the image of u, proving that u is surjective. To prove (ii), suppose m ∈ ker u. By commutativity, g um = vgm, so vgm = 0. Since v is injective, gm = 0. Thus m ∈ ker g = im f by exactness, say m = f a. Then

page 18 of Chapter 4

0 = um = uf a = f ta by commutativity. Thus ta ∈ ker f = im e by exactness. If ta = e d , then since s is surjective, we can write d = sd, so ta = e sd. By commutativity, e sd = ted, so ta = ted. By injectivity of t, a = ed. Therefore m = f a = f ed = 0 by exactness. We conclude that u is injective. ♣ 4.7.3 Corollary: The Short Five Lemma Consider the following commutative diagram with exact rows. f g 0 → A → M → B → 0 t↓ →

0

A

u↓ → f

M

v↓ → g

B

→

0

If t and v are isomorphisms, so is u. Proof. Apply the ﬁve lemma with C = D = C = D = 0, and s and w the zero maps. ♣ We can now deal with splitting of short exact sequences. 4.7.4 Proposition Let 0

→

A

f → M

g →

→

B

0

be a short exact sequence. The following conditions are equivalent, and deﬁne a split exact sequence. (i) The sequence splits on the right. (ii) The sequence splits on the left. (iii) There is an isomorphism u of M and A ⊕ B such that the following diagram is commutative. f g 0 → A → M → B → 0 || 0

→ A

u↓ → i

A⊕B

|| → π

B

→

0

Thus M is isomorphic to the direct sum of A and B, and in addition, f can be identiﬁed with the injection i of A into A ⊕ B, and g with the projection π of the direct sum onto B. (The double vertical bars indicate the identity map.) Proof. It follows from our earlier discussion of diagram (3) that (iii) implies (i) and (ii). To show that (i) implies (iii), let h be a homomorphism of B into M such that gh = 1. We claim that M = ker g ⊕ h(B). First, suppose that m ∈ M . Write m = (m − hgm) + hgm; then hgm ∈ h(B) and g(m − hgm) = gm − ghgm = gm − 1gm = gm − gm = 0. Second, suppose m ∈ ker g ∩ h(B), with m = hb. Then 0 = gm = ghb = 1b = b, so m = hb = h0 = 0, proving the claim. Now since ker g = im f by exactness, we may express any m ∈ M in the form m = f a + hb. We take um = a + b, which makes sense because both f and h are injective and f (A) ∩ h(B) = 0. This forces the diagram of (iii) to be commutative, and u is therefore an isomorphism by the short ﬁve lemma. Finally, we show that (ii) implies (iii). Let e be a homomorphism of M into A such that ef = 1. In this case, we claim that M = f (A) ⊕ ker e.

page 19 of Chapter 4

If m ∈ M then m = f em + (m − f em) and f em ∈ f (A), e(m − f em) = em − ef em = em − em = 0. If m ∈ f (A) ∩ ker e, then, with m = f a, we have 0 = em = ef a = a, so m = 0, and the claim is veriﬁed. Now if m ∈ M we have m = f a + m with a ∈ A and m ∈ ker e. We take u(m) = a + g(m ) = a + gm since gf = 0. (The deﬁnition of u is unambiguous because f is injective and f (A) ∩ ker e = 0.) The choice of u forces the diagram to be commutative, and again u is an isomorphism by the short ﬁve lemma. ♣ 4.7.5 Corollary If the sequence →

0

f →

A

g → B

M

→

0

is split exact with splitting maps e and h as in (3), then the “backwards” sequence 0 ←

e ←

A

h ← B

M

←

0

is also split exact, with splitting maps g and f . Proof. Simply note that gh = 1 and ef = 1. ♣ A device that I use to remember which way the splitting maps go (i.e., it’s ef = 1, not f e = 1) is that the map that is applied ﬁrst points inward toward the “center” M . Problems For Section 4.7 Consider the following commutative diagram with exact rows: 0

→

f →

A

B

g →

v↓ 0

→

A

→ f

B

C w↓

→ g

C

Our objective in Problems 1-3 is to ﬁnd a homomorphism u : A → A such that the square ABB A , hence the entire diagram, is commutative. 1. Show that if u exists, it is unique. 2. If a ∈ A, show that vf a ∈ im f . 3. If vf a = f a , deﬁne ua appropriately. Now consider another commutative diagram with exact rows: A

f →

w↓ A

B

g →

C

→ 0

C

→

v↓ → f

B

→ g

0

In Problems 4 and 5 we are to deﬁne u : C → C so that the diagram will commute. 4. If c ∈ C, then since g is surjective, c = gb for some b ∈ B. Write down the only possible deﬁnition of uc. 5. In Problem 4, b is not unique. Show that your deﬁnition of u does not depend on the particular b.

page 20 of Chapter 4

Problems 6-11 refer to the diagram of the short ﬁve lemma (4.7.3). Application of the four lemma is very eﬃcient, but a direct attack is also good practice. 6. 7. 8. 9.

If If If If

t and v are injective, so is u. t and v are surjective, so is u. t is surjective and u is injective, then v is injective. u is surjective, so is v. By Problems 8 and 9, if t and u are isomorphisms, so is v.

10. If u is injective, so is t. 11. If u is surjective and v is injective, then t is surjective. Note that by Problems 10 and 11, if u and v are isomorphisms, so is t. 12. If you have not done so earlier, do Problem 8 directly, without appealing to the four lemma. 13. If you have not done so earlier, do Problem 11 directly, without appealing to the four lemma.

page 1 of Chapter 5

CHAPTER 5 SOME BASIC TECHNIQUES OF GROUP THEORY 5.1 Groups Acting on Sets In this chapter we are going to analyze and classify groups, and, if possible, break down complicated groups into simpler components. To motivate the topic of this section, let’s look at the following result. 5.1.1 Cayley’s Theorem Every group is isomorphic to a group of permutations. Proof. The idea is that each element g in the group G corresponds to a permutation of the set G itself. If x ∈ G, then the permutation associated with g carries x into gx. If gx = gy, then premultiplying by g −1 gives x = y. Furthermore, given any h ∈ G, we can solve gx = h for x. Thus the map x → gx is indeed a permutation of G. The map from g to its associated permutation is injective, because if gx = hx for all x ∈ G, then (take x = 1) g = h. In fact the map is a homomorphism, since the permutation associated with hg is multiplication by hg, which is multiplication by g followed by multiplication by h, h ◦ g for short. Thus we have an embedding of G into the group of all permutations of the set G. ♣ In Cayley’s theorem, a group acts on itself in the sense that each g yields a permutation of G. We can generalize to the notion of a group acting on an arbitrary set. 5.1.2 Deﬁnitions and Comments The group G acts on the set X if for each g ∈ G there is a mapping x → gx of X into itself, such that (1)

h(gx) = (hg)x for every g, h ∈ G and x ∈ X

and (2)

1x = x for every x ∈ X.

As in (5.1.1), x → gx deﬁnes a permutation of X. The main point is that the action of g is a permutation because it has an inverse, namely the action of g −1 . (Explicitly, the inverse of x → gx is y → g −1 y.) Again as in (5.1.1), the map from g to its associated permutation Φ(g) is a homomorphism of G into the group SX of permutations of X. But we do not necessarily have an embedding. If gx = hx for all x, then in (5.1.1) we were able to set x = 1, the identity element of G, but this resource is not available in general. We have just seen that a group action induces a homomorphism from G to SX , and there is a converse assertion. If Φ is a homomorphism of G to SX , then there is a corresponding action, deﬁned by gx = Φ(g)x, x ∈ X. Condition (1) holds because Φ is a homomorphism, and (2) holds because Φ(1) must be the identity of SX . The kernel of Φ is known as the kernel of the action; it is the set of all g ∈ G such that gx = x for all x, in other words, the set of g’s that ﬁx everything in X. 5.1.3 Examples 1. (The regular action) Every group acts on itself by multiplication on the left, as in (5.1.1). In this case, the homomorphism Φ is injective, and we say that the action is faithful. [Similarly, we can deﬁne an action on the right by (xg)h = x(gh), x1 = x, and then G acts on itself by right multiplication. The problem is that Φ(gh) = Φ(h) ◦ Φ(g), an antihomomorphism. The damage can be repaired by writing function values as xf rather than f (x), or by deﬁning the action of g to be multiplication on the right by g −1 . We will avoid the diﬃculty by restricting to actions on the left.] 2. (The trivial action) We take gx = x for all g ∈ G, x ∈ X. This action is highly unfaithful. 3. (Conjugation on elements) We use the notation g • x for the action of g on x, and we set g • x = gxg −1 , called the conjugate of x by g, for g and x in the group G. Since

page 2 of Chapter 5

hgxg −1 h−1 = (hg)x(hg)−1 and 1x1−1 = x, we have a legal action of G on itself. The kernel is {x : gxg −1 = x for all x}, that is, {x : gx = xg for all x}. Thus the kernel is the set of elements that commute with everything in the group. This set is called the center of G, written Z(G). 4. (Conjugation on subgroups) If H is a subgroup of G, we take g • H = gHg −1 . Note that gHg −1 is a subgroup of G, called the conjugate subgroup of H by g, since gh1 g −1 gh2 g −1 = g(h1 h2 )g −1 and (ghg −1 )−1 = gh−1 g −1 . As in Example (3), we have a legal action of G on the set of subgroups of G. 5. (Conjugation on subsets) This is a variation of the previous example. In this case we let G act by conjugation on the collection of all subsets of G, not just subgroups. The veriﬁcation that the action is legal is easier, because gHg −1 is certainly a subset of G. 6. (Multiplication on left cosets) Let G act on the set of left cosets of a ﬁxed subgroup G by g • (xH) = (gx)H. By deﬁnition of set multiplication, we have a legitimate action. 7. (Multiplication on subsets) Let G act on all subsets of G by g • S = gS = {gx : x ∈ S}. Again the action is legal by deﬁnition of set multiplication. Problems For Section 5.1 1. Let G act on left cosets of H by multiplication, as in Example 6. Show that the kernel of the action is a subgroup of H. 2. Suppose that H is a proper subgroup of G of index n, and that G is a simple group, that is, G has no normal subgroups except G itself and {1}. Show that G can be embedded in Sn . 3. Suppose that G is an inﬁnite simple group. Show that every for every subgroup H of G, the index [G : H] is inﬁnite. 4. Let G act on left cosets of H by multiplication. Show that the kernel of the action is N=

xHx−1 .

x∈G

5. Continuing Problem 4, if K is any normal subgroup of G contained in H, show that K ≤ N . Thus N is the largest normal subgroup of G contained in H; N is called the core of H in G. 6. Here is some extra practice with left cosets of various subgroups. Let H and K be subgroups of G, and consider the map f which assigns to the coset g(H ∩ K) the pair of cosets (gH, gK). Show that f is well-deﬁned and injective, and therefore [G : H ∩ K] ≤ [G : H][G : K]. Thus (Poincar´e ) the intersection of ﬁnitely many subgroups of ﬁnite index also has ﬁnite index. 7. If [G : H] and [G : K] are ﬁnite and relatively prime, show that the inequality in the preceding problem is actually an equality. 8. Let H be a subgroup of G of ﬁnite index n, and let G act on left cosets xH by multiplication. Let N be the kernel of the action, so that N H by Problem 1. Show that [G : N ] divides n!. 9. Let H be a subgroup of G of ﬁnite index n > 1. If |G| does not divide n!, show that G is not simple. 5.2 The Orbit-Stabilizer Theorem

page 3 of Chapter 5

5.2.1 Deﬁnitions and Comments Suppose that the group G acts on the set X. If we start with the element x ∈ X and successively apply group elements in all possible ways, we get B(x) = {gx : g ∈ G} which is called the orbit of x under the action of G. The action is transitive (we also say that G acts transitively on X) if there is only one orbit, in other words, for any x, y ∈ X, there exists g ∈ G such that gx = y. Note that the orbits partition X, because they are the equivalence classes of the equivalence relation given by y ∼ x iﬀ y = gx for some g ∈ G. The stabilizer of an element x ∈ X is G(x) = {g ∈ G : gx = x}, the set of elements that leave x ﬁxed. A direct veriﬁcation shows that G(x) is a subgroup. This is a useful observation because any set that appears as a stabilizer in a group action is guaranteed to be a subgroup; we need not bother to check each time. Before proceeding to the main theorem, let’s return to the examples considered in (5.1.3). 5.2.2 Examples 1. The regular action of G on G is transitive, and the stabilizer of x is the subgroup {1}. 2. The trivial action is not transitive (except in trivial cases), in fact, B(x) = {x} for every x. The stabilizer of x is the entire group G. 3. Conjugation on elements is not transitive (see Problem 1). The orbit of x is the set of conjugates gxg −1 of x, that is, B(x) = {gxg −1 : g ∈ G}, which is known as the conjugacy class of x. The stabilizer of x is G(x) = {g : gxg −1 = x} = {g : gx = xg}, the set of group elements that commute with x. This set is called the centralizer of x, written CG (x). Similarly, the centralizer CG (S) of an arbitrary subset S ⊆ G is deﬁned as the set of elements of G that commute with everything in S. (Here, we do need to check that CG (S) is a subgroup, but there is no problem.) 4. Conjugation on subgroups is not transitive. The orbit of H is {gHg −1 : g ∈ G}, the collection of conjugate subgroups of H. The stabilizer of H is {g : gHg −1 = H}, which is called the normalizer of H, written NG (H). If K is a subgroup of G containing H, we have H K iﬀ gHg −1 = H for every g ∈ K and this holds iﬀ K is a subgroup of NG (H). Thus NG (H) is the largest subgroup of G in which H is normal. 5. Conjugation on subsets is not transitive, and the orbit of the subset S is {gSg −1 : g ∈ G}. The stabilizer of S is the normalizerNG (S) = {g : gSg −1 = S}. 6. Multiplication on left cosets is transitive; a solution of g(xH) = yH for x is x = g −1 y. The stabilizer of xH is {g : gxH = xH} = {g : x−1 gx ∈ H} = {g : g ∈ xHx−1 } = xHx−1 ,

page 4 of Chapter 5

the conjugate of H by x. Taking x = 1, we see that the stabilizer of H is H itself. 7. Multiplication on subsets is not transitive. The stabilizer of S is {g : gS = S}, the set of elements of G that permute the elements of S. 5.2.3 The Orbit-Stabilizer Theorem Suppose that a group G acts on a set X. Let B(x) be the orbit of x ∈ X, and let G(x) be the stabilizer of x. Then the size of the orbit is the index of the stabilizer, that is, |B(x)| = [G : G(x)]. Thus if G is ﬁnite, then |B(x)| = |G|/|G(x)|; in particular, the orbit size divides the order of the group. Proof. If y belongs to the orbit of x, say y = gx. We take f (y) = gH, where H = G(x) is the stabilizer of x. To check that f is a well-deﬁned map of B(x) to the set of left cosets of H, let y = g1 x = g2 x. Then g2−1 g1 x = x, so g2−1 g1 ∈ H, i.e., g1 H = g2 H. Since g is an arbitrary element of G, f is surjective. If g1 H = g2 H, then g2−1 g1 ∈ H, so that g2−1 g1 x = x, and consequently g1 x = g2 x. Thus if y1 = g1 x, y2 = g2 x, and f (y1 ) = f (y2 ), then y1 = y2 , proving f injective. ♣ Referring to (5.2.2), Example 3, we see that B(x) is an orbit of size 1 iﬀ x commutes with every g ∈ G, i.e., x ∈ Z(G), the center of G. Thus if G is ﬁnite and we select one element xi from each conjugacy class of size greater than 1, we get the class equation |G| = |Z(G)| + [G : CG (xi )]. i

We know that a group G acts on left cosets of a subgroup K by multiplication. To prepare for the next result, we look at the action of a subgroup H of G on left cosets of K. Since K is a left coset of K, it has an orbit given by {hK : h ∈ H}. The union of the sets hK is the set product HK. The stabilizer of K is not K itself, as in Example 6; it is {h ∈ H : hK = K}. But hK = K(= 1K) if and only if h ∈ K, so the stabilizer is H ∩ K. 5.2.4 Proposition If H and K are subgroups of the ﬁnite group G, then |HK| =

|H||K| . |H K|

Proof. The cosets in the orbit of K are disjoint, and each has |K| members. Since, as remarked above, the union of the cosets is HK, there must be exactly |HK|/|K| cosets in the orbit. Since the index of the stabilizer of K is |H/H ∩ K|, the result follows from the orbit-stabilizer theorem. ♣ Problems For Section 5.2 1. Let σ be the permutation (1,2,3,4,5) and π the permutation (1,2)(3,4). Then πσπ −1 , the conjugate of σ by π, can be obtained by applying π to the symbols of σ to get (2,1,4,3,5). Re−1 versing the process, if we are givenτ = (1, 2)(3, 4) and we specify that µτ µ = (1, 3)(2, 5), 1 2 3 4 5 we can take µ = . This suggests that two permutations are conjugate if 1 3 2 5 4 and only if they have the same cycle structure. Explain why this works. 2. Show that if S is any subset of G, then the centralizer of S is a normal subgroup of the normalizer of S. (Let the normalizer NG (S) act on S by conjugation on elements.) 3. Let G(x) be the stabilizer of x under a group action. Show that stabilizers of elements in the orbit of x are conjugate subgroups. Explicitly, for every g ∈ G and x ∈ X we have G(gx) = gG(x)g −1 .

page 5 of Chapter 5

4. Let G act on the set X. Show that Ψ(gG(x)) = gx is a well-deﬁned injective mapping of G/G(x) into X, and is bijective if the action is transitive. 5. Continuing Problem 4, let G act transitively on X, and choose any x ∈ X. Show that the action of G on X is essentially the same as the action of G on the left cosets of the stabilizer subgroup G(x). This is the meaning of the assertion that“any transitive G-set is isomorphic to a space of left cosets”. Give an appropriate formal statement expressing this idea. 6. Suppose that G is a ﬁnite group, and for every x, y ∈ G such that x = 1 and y = 1, x and y are conjugate. Show that the order of G must be 1 or 2. 7. First note that if r is a positive rational number and k a ﬁxed positive integer, there are only ﬁnitely many positive integer solutions of the equation 1 1 + ··· + = r. x1 xk Outline of proof: If xk is the smallest xi , the left side is at most k/xk , so 1 ≤ xk ≤ k/r and there are only ﬁnitely many choices for xk . Repeat this argument for the equation 1 1 1 x1 + · · · + xk−1 = r − xk . Now set r = 1 and let N (k) be an upper bound on all the xi ’s in all possible solutions. If G is a ﬁnite group with exactly k conjugacy classes, show that the order of G is at most N (k). 5.3 Application To Combinatorics The theory of group actions can be used to solve a class of combinatorial problems. To set up a typical problem, consider the regular hexagon of Figure 5.3.1, and recall the dihedral group D12 , the group of symmetries of the hexagon (Section 1.2).

3

2 1

4 6

5

Figure 5.3.1 If R is rotation by 60 degrees and F is reﬂection about the horizontal line joining vertices 1 and 4, the 12 members of the group may be listed as follows. I = identity,

R = (1, 2, 3, 4, 5, 6),

R3 = (1, 4)(2, 5)(3, 6), F = (2, 6)(3, 5),

R2 = (1, 3, 5)(2, 4, 6),

R4 = (1, 5, 3)(2, 6, 4),

RF = (1, 2)(3, 6)(4, 5),

R3 F = (1, 4)(2, 3)(5, 6),

R5 = (1, 6, 5, 4, 3, 2)

R2 F = (1, 3)(4, 6)

R4 F = (1, 5)(2, 4),

R5 F = (1, 6)(2, 5)(3, 4).

(As before, RF means F followed by R.) Suppose that we color the vertices of the hexagon, and we have n colors available (we are not required to use every color). How many distinct colorings are there? Since we may choose the color of any vertex in n ways, a logical answer is n6 . But this answer does not describe the physical situation accurately. To see what is happening, suppose we have two colors, yellow (Y ) and blue (B). Then the coloring C1 =

1 B

2 B

3 Y

4 Y

5 Y

6 B

page 6 of Chapter 5

is mapped by RF to C2 =

1 B

2 B

3 B

4 Y

5 Y

6 Y

(For example, vertex 3 goes to where vertex 6 was previously, delivering the color yellow to vertex 6.) According to our counting scheme, C2 is not the same as C1 . But imagine that we have two rigid necklaces in the form of a hexagon, one colored by C1 and the other by C2 . If both necklaces were lying on a table, it would be diﬃcult to argue that they are essentially diﬀerent, since one can be converted to a copy of the other simply by ﬂipping it over and then rotating it. Let’s try to make an appropriate mathematical model. Any group of permutations of a set X acts on X in the natural way: g • x = g(x). In particular, the dihedral group G acts on the vertices of the hexagon, and therefore on the set S of colorings of the vertices. The above discussion suggests that colorings in the same orbit should be regarded as equivalent, so the number of essentially diﬀerent colorings is the number of orbits. The following result will help us do the counting. 5.3.1 Orbit-Counting Theorem Let the ﬁnite group G act on the ﬁnite set X, and let f (g) be the number of elements of X ﬁxed by g, that is, the size of the set {x ∈ X : g(x) = x}. Then the number of orbits is 1 f (g), |G| g∈G

the average number of points left ﬁxed by elements of G. Proof. We use a standard combinatorial technique called “counting two ways”. Let T be the set of all ordered pairs (g, x) such that g ∈ G, x ∈ X, and gx = x. For any x ∈ X, the number of g’s such that (g, x) ∈ T is the size of the stabilizer subgroup G(x), hence

|T | =

|G(x)|.

(1)

x∈X

Now for any g ∈ G, the number of x’s such that (g, x) ∈ T is f (g), the number of ﬁxed points of g. Thus |T | = f (g). (2) g∈G

Divide (1) and (2) by the order of G to get |G(x)| 1 f (g). = |G| |G|

x∈X

(3)

g∈G

But by the orbit-stabilizer theorem (5.2.3), |G|/|G(x)| is |B(x)|,the size of the orbit of x. If, for example, an orbit has 5 members, then 1/5 will appear 5 times in the sum on the left side of (3), for a total contribution of 1. Thus the left side of (3) is the total number of orbits. ♣ We can now proceed to the next step in the analysis. 5.3.2 Counting the Number of Colorings Fixed by a Given Permutation Let π = R2 = (1, 3, 5)(2, 4, 6). Since π(1) = 3 and π(3) = 5, vertices 1,3 and 5 have the same color. Similarly, vertices 2,4 and 6 must have the same color. If there are n colors available, we can choose the color of each cycle in n ways, and the total number of choices is n2 . If π = F = (2, 6)(3, 5), then as before we choose 1 color out of n for each cycle, but in this case we still have to color the vertices 1 and 4. Here is a general statement that covers both situations.

page 7 of Chapter 5

If π has c cycles, counting cycles of length 1, then the number of colorings ﬁxed by π is nc . To emphasize the need to consider cycles of length 1, we can write F as (2,6)(3,5)(1)(4). From the cycle decompositions given at the beginning of the section, we have one permutation (the identity) with 6 cycles, three with 4 cycles, four with 3 cycles, two with 2 cycles, and two with 1 cycle. Thus the number of distinct colorings is 1 6 (n + 3n4 + 4n3 + 2n2 + 2n). 12 5.3.3 A Variant We now consider a slightly diﬀerent question. How many distinct colorings of the vertices of a regular hexagon are there if we are forced to color exactly three vertices blue and three vertices yellow? The group G is the same as before, but the set S is diﬀerent. Of the 64 possible colorings of the vertices, only 63 = 20 are legal, since 3 vertices out of 6 are chosen to be colored blue; the other vertices must be colored yellow. If π is a permutation of G, then within each cycle of π, all vertices have the same color, but in contrast to the previous example, we do not have a free choice of color for each cycle. To see this, consider R2 = (1, 3, 5)(2, 4, 6). The cycle (1,3,5) can be colored blue and (2,4,6) yellow, or vice versa, but it is not possible to color all six vertices blue, or to color all vertices yellow. Thus f (R2 ) = 2. If π = F = (2, 6)(3, 5)(1)(4), a ﬁxed coloring is obtained by choosing one of the cycles of length 2 and one of the cycles of length 1 to be colored blue, thus producing 3 blue vertices. Consequently, f (F ) = 4. To obtain f (I), note that all legal colorings are ﬁxed by I, so f (I) is the number of colorings of 6 vertices with exactly 3 blue and 3 yellow vertices, namely, 63 = 20. From the cycle decompositions of the members of G, there are two permutations with f = 2, three with f = 4, and one with f = 20; the others have f = 0. Thus the number of distinct colorings is 1 (2(2) + 3(4) + 20) = 3. 12 Problems For Section 5.3 1. Assume that two colorings of the vertices of a square are equivalent if one can be mapped into the other by a permutation in the dihedral group G = D8 . If n colors are available, ﬁnd the number of distinct colorings. 2. In Problem 1, suppose that we color the sides of the square rather than the vertices. Do we get the same answer? 3. In Problem 1, assume that only two colors are available, white and green. There are 16 unrestricted colorings, but only 6 equivalence classes. List the equivalence classes explicitly. 4. Consider a rigid rod lying on the x-axis from x = −1 to x = 1, with three beads attached. The beads are located at the endpoints (-1,0) and (1,0), and at the center (0,0). The beads are to be painted using n colors, and two colorings are regarded as equivalent if one can be mapped into the other by a permutation in the group G = {I, σ}, where σ is the 180 degree rotation about the vertical axis. Find the number of distinct colorings. 5. In Problem 4, ﬁnd the number of distinct colorings if the color of the central bead is always black. 6. Consider the group of rotations of the regular tetrahedron (see Figure 5.3.2); G consists of the following permutations. (i) The identity; (ii) Rotations by 120 degrees, clockwise or counterclockwise, about an axis through a vertex and the opposite face. There are 8 such rotations (choose 1 of 4 vertices, then choose a clockwise or counterclockwise direction); (iii) Rotations by 180 degrees about the line joining the midpoints of two nontouching edges. There are 3 such rotations.

page 8 of Chapter 5

Argue geometrically to show that there are no other rotations in the group, and show that G is isomorphic to the alternating group A4 .

3

4

2

1 Figure 5.3.2 7. Find the number of distinct colorings of the vertices of a regular tetrahedron, if colorings that can be rotated into each other are equivalent. 8. In Problem 7, assume that n = 4 and label the colors B,Y,W,G. Find the number of distinct colorings if exactly two vertices must be colored B. 9. The group G of rotations of a cube consists of the following permutations of the faces. (i) The identity. (ii) Rotations of ±90 or 180 degrees about a line through the center of two opposite faces; there are 3 × 3 = 9 such rotations. (iii) Rotations of ±120 degrees about a diagonal of the cube, i.e., a line joining two opposite vertices (vertices that are a maximal distance apart). There are 4 diagonals, so there are 4 × 2 = 8 such rotations. (iv) Rotations of 180 degrees about a line joining the midpoints of two opposite edges. There are 6 such rotations. (An axis of rotation is determined by selecting one of the four edges on the bottom of the cube, or one of the two vertical edges on the front face.) Argue geometrically to shows that there are no other rotations in the group, and show that G is isomorphic to the symmetric group S4 . 10. If six colors are available and each face of a cube is painted a diﬀerent color, ﬁnd the number of distinct colorings. 11. Let G be the group of rotations of a regular p-gon, where p is an odd prime. If the vertices of the p-gon are to be painted using n colors, ﬁnd the number of distinct colorings. 12. Use the result of Problem 11 to give an unusual proof of Fermat’s little theorem. 5.4 The Sylow Theorems Considerable information about the structure of a ﬁnite group G can be obtained by factoring the order of G. Suppose that |G| = pr m where p is prime, r is a positive integer, and p does not divide m. Then r is the highest power of p that divides the order of G. We will prove, among other things, that G must have a subgroup of order pr , and any two such subgroups must be conjugate. We will need the following result about binomial coeﬃcients. 5.4.1 Lemma If n = pr m where p is prime, then pnr ≡ m mod p. Thus if p does not r divide m, then it does not divide pprm . Proof. By the binomial expansion modulo p (see Section 3.4), we have r

r

r

r

(X + 1)p ≡ X p + 1p = X p + 1 mod p. Raise both sides to the power m to obtain r

(X + 1)n ≡ (X p + 1)m mod p.

page 9 of Chapter 5 r

On the left side, the coeﬃcient of X p is result follows. ♣

n pr

, and on the right side, it is

m m−1

= m. The

5.4.2 Deﬁnitions and Comments Let p be a prime number. The group G is said to be a p-group if the order of each element of G is a power of p. (The particular power depends on the element.) If G is a ﬁnite group, then G is a p-group iﬀ the order of G is a power of p. [The “if” part follows from Lagrange’s theorem, and the “only if” part is a corollary to the Sylow theorems; see (5.4.5).] If |G| = pr m, where p does not divide m, then a subgroup P of G of order pr is called a Sylow p-subgroup of G. Thus P is a p-subgroup of G of maximum possible size. 5.4.3 The Sylow Theorems Let G be a ﬁnite group of order pr m, where p is prime, r is a positive integer, and p does not divide m. Then (1) G has at least one Sylow p-subgroup, and every p-subgroup of G is contained in a Sylow p-subgroup. (2) Let np be the number of Sylow p-subgroups of G. Then np ≡ 1 mod p and np divides m. (3) All Sylow p-subgroups are conjugate. Thus if we deﬁne an equivalence relation on subgroups by H ∼ K iﬀ H = gKg −1 for some g ∈ G, then the Sylow p-subgroups comprise a single equivalence class. [Note that the conjugate of a Sylow p-subgroup is also a Sylow p-subgroup, since it has the same number of elements pr .] Proof. (1)Let G act on subsets of G of size pr by left multiplication. The number of such subsets r is pprm , which is not divisible by p by (5.4.1). Consequently, there is at least one subset S whose orbit size is not divisible by p. If P is the stabilizer of S, then by (5.2.3), the size of the orbit is [G : P ] = |G|/|P | = pr m/|P |. For this to fail to be divisible by p, we must have pr ||P |, and therefore pr ≤ |P |. But for any ﬁxed x ∈ S, the map of P into S given by g → gx is injective. (The map is indeed into S because g belongs to the stabilizer of S, so that gS = S.) Thus |P | ≤ |S| = pr . We conclude that |P | = pr , hence P is a Sylow p-subgroup. So far, we have shown that a Sylow p-subgroup P exists, but not that every p-subgroup is contained in a Sylow p-subgroup. We will return to this in the course of proving (2) and (3). (2) and (3) Let X be the set of all Sylow p-subgroups of G. Then |X| = np and P acts on X by conjugation, i.e., g • Q = gQg −1 , g ∈ P . By (5.2.3), the size of any orbit divides |P | = pr , hence is a power of p. Suppose that there is an orbit of size 1, that is, a set Q ∈ X such that gQg −1 = Q for every g ∈ P . (There is at least one such set, namely P .) We would like to say that Q P , but Q need not be a subgroup of P . However, we can conclude that Q is a normal subgroup of < P, Q >, the subgroup generated by P and Q (see the proof of (1.4.7)). By (1.4.3), P Q(= QP ) is a subgroup of < P, Q >. Since |P | = |Q| = pr it follows from (5.2.4) that |P Q| is a power of p, say pc . We must have c ≤ r because P Q is a subgroup of G (hence |P Q| divides |G|). Thus pr = |P | ≤ |P Q| ≤ pr , so |P | = |P Q| = pr . But P is a subset of P Q, and since all sets are ﬁnite, we conclude that P = P Q, and therefore Q ⊆ P . Since both P and Q are of size pr , we have P = Q. Thus there is only one orbit of size 1, namely {P }. Since by (5.2.3), all other orbit sizes are of the form pc where c ≥ 1, it follows that np ≡ 1 mod p. Now let R be a p-subgroup of G, and let R act by multiplication on Y , the set of left cosets of P . Since |Y | = [G : P ] = |G|/|P | = pr m/pr = m, p does not divide |Y |. Therefore

page 10 of Chapter 5

some orbit size is not divisible by p. By (5.2.3), every orbit size divides |R|, hence is a power of p. (See (5.4.5) below. We are not going around in circles because (5.4.4) and (5.4.5) only depend on the existence of Sylow subgroups, which we have already established.) Thus there must be an orbit of size 1, say {gP } with g ∈ G. If h ∈ R then hgP = gP , that is, g −1 hg ∈ P , or equally well, h ∈ gP g −1 . Consequently, R is contained in a conjugate of P . If R is a Sylow p-subgroup to begin with, then R is a conjugate of P , completing the proof of (1) and (3). To ﬁnish (2), we must show that np divides m. Let G act on subgroups by conjugation. The orbit of P has size np by (3), so by (5.2.3), np divides |G| = pr m. But p cannot be a prime factor of np , since np ≡ 1 mod p. It follows that np must divide m. ♣ 5.4.4 Corollary (Cauchy’s Theorem) If the prime p divides the order of G, then G has an element of order p. Proof. Let P be a Sylow p-subgroup of G, and pick x ∈ P with x = 1. The order of x is a k−1 power of p, say |x| = pk . Then xp has order p. ♣ 5.4.5 Corollary The ﬁnite group G is a p-group if and only if the order of G is a power of p. Proof. If the order of G is not a power of p, then it is divisible by some other prime q. But in this case, G has a Sylow q-subgroup, and therefore by (5.4.4), an element of order q. Thus G cannot be a p-group. The converse was done in (5.4.2). ♣ Problems For Section 5.4 1. Under the hypothesis of the Sylow theorems, show that G has a subgroup of index np . 2. Let P be a Sylow p-subgroup of the ﬁnite group G, and let Q be any p-subgroup. If Q is contained in the normalizer NG (P ), show that P Q is a p-subgroup. 3. Continuing Problem 2, show that Q is contained in P . 4. Let P be a Sylow p-subgroup of the ﬁnite group G, and let H be a subgroup of G that contains the normalizer NG (P ). (a) If g ∈ NG (H), show that P and gP g −1 are Sylow p-subgroups of H, hence they are conjugate in H. (b) Show that NG (H) = H. 5. Let P be a Sylow p-subgroup of the ﬁnite group G, and let N be a normal subgroup of G. Assume that p divides |N | and |G/N |, so that N and G/N have Sylow p-subgroups. Show that [P N : P ] and p are relatively prime, and then show that P ∩ N is a Sylow p-subgroup of N . 6. Continuing Problem 5, show that P N/N is a Sylow p-subgroup of G/N . 7. Suppose that P is the unique Sylow P -subgroup of G. [Equivalently, P is a normal Sylow p-subgroup of G; see (5.5.4).] Show that for each automorphism f of G, we have f (P ) = P . [Thus P is a characteristic subgroup of G; see (5.7.1).] 8. The Sylow theorems are about subgroups whose order is a power of a prime p. Here is a result about subgroups of index p. Let H be a subgroup of the ﬁnite group G, and assume that [G : H] = p. Let N be a normal subgroup of G such that N ≤ H and [G : N ] divides p! (see Section 5.1, Problem 8). Show that [H : N ] divides (p − 1)!. 9. Continuing Problem 8, let H be a subgroup of the ﬁnite group G, and assume that H has index p, where p is the smallest prime divisor of |G|. Show that H G. 5.5 Applications Of The Sylow Theorems The Sylow theorems are of considerable assistance in the problem of classifying, up to isomorphism, all ﬁnite groups of a given order n. But in this area, proofs tend to involve intricate combinatorial arguments, best left to specialized texts in group theory. We will try to illustrate some of the basic ideas while keeping the presentation clean and crisp.

page 11 of Chapter 5

5.5.1 Deﬁnitions and Comments A group G is simple if G = {1} and the only normal subgroups of G are G itself and {1}. We will see later that simple groups can be regarded as building blocks for arbitrary ﬁnite groups. Abelian simple groups are already very familiar to us; they are the cyclic groups of prime order. For if x ∈ G, x = 1, then by simplicity (and the fact that all subgroups of an abelian group are normal), G =< x >. If G is not of prime order, then G has a nontrivial proper subgroup by (1.1.4), so G cannot be simple. The following results will be useful. 5.5.2 Lemma If H and K are normal subgroups of G and the intersection of H and K is trivial (i.e., {1}), then hk = kh for every h ∈ H and k ∈ K. Proof. We did this in connection with direct products; see the beginning of the proof of (1.5.2). ♣ 5.5.3 Proposition If P is a nontrivial ﬁnite p-group, then P has a nontrivial center. Proof. Let P act on itself by conjugation; see (5.1.3) and (5.2.2), Example 3. The orbits are the conjugacy classes of P . The element x belongs to an orbit of size 1 iﬀ x is in the center Z(P ), since gxg −1 = x for all g ∈ P iﬀ gx = xg for all g ∈ P iﬀ x ∈ Z(P ). By the orbit-stabilizer theorem, an orbit size that is greater than 1 must divide |P |, and therefore must be a positive power of p. If Z(P ) = {1}, then we have one orbit of size 1, with all other orbit sizes ≡ 0 mod p. Thus |P | ≡ 1 mod p, contradicting the assumption that P is a nontrivial p-group. ♣ 5.5.4 Lemma P is a normal Sylow p-subgroup of G if and only if P is the unique Sylow p-subgroup of G. Proof. By Sylow (3), the Sylow p-subgroups form a single equivalence class of conjugate subgroups. This equivalence class consists of a single element {P } iﬀ gP g −1 = P for every g ∈ G , that is, iﬀ P G. ♣ 5.5.5 Proposition Let G be a ﬁnite, nonabelian simple group. If the prime p divides the order of G, then the number np of Sylow p-subgroups of G is greater than 1. Proof. If p is the only prime divisor of |G|, then G is a nontrivial p-group, hence Z(G) is nontrivial by (5.5.3). Since Z(G) G (see (5.1.3, Example 3), Z(G) = G, so that G is abelian, a contradiction. Thus |G| is divisible by at least two distinct primes, so if P is a Sylow p-subgroup, then {1} 1. ♣ We can now derive some properties of groups whose order is the product of two distinct primes. 5.5.6 Proposition Let G be a group of order pq, where p and q are distinct primes. (i) If q ≡ 1 mod p, then G has a normal Sylow p-subgroup. (ii) G is not simple. (iii) If p ≡ 1 mod q and q ≡ 1 mod p, then G is cyclic. Proof. (i) By Sylow (2), np ≡ 1 mod p and np |q, so np = 1. The result follows from (5.5.4). (ii) We may assume without loss of generality that p > q. Then p cannot divide q − 1, so q ≡ 1 mod p. By (i), G has a normal Sylow p-subgroup, so G is not simple. (iii) By (i), G has a normal Sylow p-subgroup P and a normal Sylow q-subgroup Q. Since P and Q are of prime order (p and q, respectively), they are cyclic. If x generates P and y generates Q, then xy = yx by (5.5.2). [P and Q have trivial intersection because any member of the intersection has order dividing both p and q.] But then xy has order pq = |G| (see Section 1.1, Problem 8). Thus G =< xy >. ♣

page 12 of Chapter 5

We now look at the more complicated case |G| = p2 q. The combinatorial argument in the next proof is very interesting. 5.5.7 Proposition Suppose that the order of the ﬁnite group G is p2 q, where p and q are distinct primes. Then G has either a normal Sylow p-subgroup or a normal Sylow q-subgroup. Thus G is not simple. Proof. If the conclusion is false then np and nq are both greater than 1. By Sylow (2), nq divides p2 , so nq = p or p2 , and we will show that the second case leads to a contradiction. A Sylow q-subgroup Q is of order q and is therefore cyclic. Furthermore, every element of Q except the identity is a generator of Q. Conversely, any element of order q generates a Sylow q-subgroup. Since the only divisors of q are 1 and q, any two distinct Sylow q-subgroups have trivial intersection. Thus the number of elements of G of order q is exactly nq (q − 1). If nq = p2 , then the number of elements that are not of order q is p2 q − p2 (q − 1) = p2 . Now let P be any Sylow p-subgroup of G. Then |P | = p2 , so no element of P can have order q (the orders must be 1, p or p2 ). Since there are only p2 elements of order unequal to q available, P takes care of all of them. Thus there cannot be another Sylow p-subgroup, so np = 1, a contradiction. We conclude that nq must be p. Now by Sylow (2), nq ≡ 1 mod q, hence p ≡ 1 mod q, so p > q. But np divides q, a prime, so np = q. Since np ≡ 1 mod p, we have q ≡ 1 mod p, and consequently q > p. Our original assumption that both np and nq are greater than one has led inexorably to a contradiction. ♣ Problems For Section 5.5 1. Show that every group of order 15 is cyclic. 2. If G/Z(G) is cyclic, show that G = Z(G), and therefore G is abelian. 3. Show that every group of order p2 is abelian. 4. Let G be a group with |G| = pqr, where p, q and r are distinct primes and (without loss of generality) p > q > r. Show that |G| ≥ 1 + np (p − 1) + nq (q − 1) + nr (r − 1). 5. Continuing Problem 4, if G is simple, show that np , nq and nr are all greater than 1. Then show that np = qr, nq ≥ p and nr ≥ q. 6. Show that a group whose order is the product of three distinct primes is not simple. 7. Let G be a simple group of order pr m, where r ≥ 1, m > 1, and the prime p does not divide m. Let n = np be the number of Sylow p-subgroups of G. If H = NG (P ), where P is a Sylow p-subgroup of G, then [G : H] = n (see Problem 1 of Section 5.4). Show that P cannot be normal in G (hence n > 1), and conclude that |G| must divide n!. 8. If G is a group of order 250, 000 = 24 56 , show that G is not simple. 5.6 Composition Series 5.6.1 Deﬁnitions and Comments One way to break down a group into simpler components is via a subnormal series 1 = G0 G1 · · · Gr = G. “Subnormal” means that each subgroup Gi is normal in its successor Gi+1 . In a normal series, the Gi are required to be normal subgroups of the entire group G. For convenience, the trivial subgroup {1} will be written as 1. Suppose that Gi is not a maximal normal subgroup of Gi+1 , equivalently (by the correspondence theorem) Gi+1 /Gi is not simple. Then the original subnormal series can be reﬁned by inserting a group H such that Gi ) H ) Gi+1 . We can continue reﬁning in the hope that the process will terminate (it always will if G is ﬁnite). If all factors Gi+1 /Gi are simple, we say that the group G has a composition series. [By convention, the trivial group has a composition series, namely {1} itself.]

page 13 of Chapter 5

The Jordan-H¨ older theorem asserts that if G has a composition series, the resulting composition length r and the composition factors Gi+1 /Gi are unique (up to isomorphism and rearrangement). Thus all reﬁnements lead to essentially the same result. Simple groups therefore give important information about arbitrary groups; if G1 and G2 have diﬀerent composition factors, they cannot be isomorphic. Here is an example of a composition series. Let S4 be the group of all permutations of {1, 2, 3, 4}, and A4 the subgroup of even permutations (normal in S4 by Section 1.3, Problem 6). Let V be the four group (Section 1.2, Problem 6; normal in A4 , in fact in S4 , by direct veriﬁcation). Let Z2 be any subgroup of V of order 2. Then 1 ) Z2 ) V ) A4 ) S4 . The proof of the Jordan-H¨ older theorem requires some technical machinery. 5.6.2 Lemma (i) If K H ≤ G and f is a homomorphism on G, then f (K) f (H). (ii) If K H ≤ G and N G, then N K N H. (iii) If A, B, C and D are subgroups of G with A B and C D, then A(B∩C) A(B∩D), and by symmetry, C(D ∩ A) C(D ∩ B). (iv) In (iii), A(B ∩ C) ∩ B ∩ D = C(D ∩ A) ∩ D ∩ B. Equivalently, A(B ∩ C) ∩ D = C(D ∩ A) ∩ B. Proof. (i) f (h)f (k)f (h)−1 = f (hkh−1 ) ∈ f (K). (ii) Let f be the canonical map of G onto G/N . By (i) we have N K/N N H/N . The result follows from the correspondence theorem. (iii) Apply (ii) with G = B, N = A, K = B ∩ C, H = B ∩ D. (iv) The two versions are equivalent because A(B ∩ C) ≤ B and C(D ∩ A) ≤ D. If x belongs to the set on the left, then x = ac for some a ∈ A, c ∈ B ∩ C, and x also belongs to D. But x = c(c−1 ac) = ca∗ for some a∗ ∈ A B. Since x ∈ D and c ∈ C ≤ D, we have a∗ ∈ D, hence a∗ ∈ D ∩ A. Thus x = ca∗ ∈ C(D ∩ A), and since x = ac, with a ∈ A ≤ B and c ∈ B ∩ C ≤ B, x ∈ C(D ∩ A) ∩ B. Therefore the left side is a subset of the right side, and a symmetrical argument completes the proof. ♣ The diagram below is helpful in visualizing the next result. B A

| D | | C

To keep track of symmetry, take mirror images about the dotted line. Thus the group A will correspond to C, B to D, A(B ∩ C) to C(D ∩ A), and A(B ∩ D) to C(D ∩ B). 5.6.3 Zassenhaus Lemma Let A, B, C and D be subgroups of G, with A B and C D. Then A(B ∩ D) ∼ C(D ∩ B) . = A(B ∩ C) C(D ∩ A) Proof. By part (iii) of (5.6.2), the quotient groups are well-deﬁned. An element of the group on the left is of the form ayA(B ∩ C), a ∈ A, y ∈ B ∩ D. But ay = y(y −1 ay) = ya∗ , a∗ ∈ A. Thus ayA(B ∩ C) = ya∗ A(B ∩ C) = yA(B ∩ C). Similarly, an element of the right side is of the form zC(D ∩ A) with z ∈ D ∩ B = B ∩ D. Thus if y, z ∈ B ∩ D, then yA(B ∩ C) = zA(B ∩ C) iﬀ z −1 y ∈ A(B ∩ C) ∩ B ∩ D

page 14 of Chapter 5

and by part (iv) of (5.6.2), this is equivalent to z −1 y ∈ C(D ∩ A) ∩ D ∩ B iﬀ yC(D ∩ A) = zC(D ∩ A). Thus if h maps yA(B ∩ C) to yC(D ∩ A), then h is a well-deﬁned bijection from the left to the right side of Zassenhaus’ equation. By deﬁnition of multiplication in a quotient group, h is an isomorphism. ♣ 5.6.4 Deﬁnitions and Comments If a subnormal series is reﬁned by inserting H between Gi and Gi+1 , let us allow H to coincide with Gi or Gi+1 . If all such insertions are strictly between the “endgroups”, we will speak of a proper reﬁnement. Two series are equivalent if they have the same length and their factor groups are the same, up to isomorphism and rearrangement. 5.6.5 Schreier Reﬁnement Theorem Let 1 = H0 H1 · · · Hr = G and 1 = K0 K1 · · · Ks = G be two subnormal series for the group G. Then the series have equivalent reﬁnements. Proof. Let Hij = Hi (Hi+1 ∩ Kj ), Kij = Kj (Kj+1 ∩ Hi ). By Zassenhaus we have Hi,j+1 ∼ Ki+1,j . = Hij Kij (In (5.6.3) take A = Hi , B = Hi+1 , C = Kj , D = Kj+1 ). We can now construct equivalent reﬁnements; the easiest way to see this is to look at a typical concrete example. The ﬁrst reﬁnement will have r blocks of length s, and the second will have s blocks of length r. Thus the length will be rs in both cases. With r = 2 and s = 3, we have 1 = H00 H01 H02 H03 = H1 = H10 H11 H12 H13 = H2 = G, 1 = K00 K10 K20 = K1 = K01 K11 K21 = K2 = K02 K12 K22 = K3 = G. The corresponding factor groups are H01 /H00 ∼ = K10 /K00 , H02 /H01 ∼ = K11 /K01 , H03 /H02 ∼ = K12 /K02 H11 /H10 ∼ = K20 /K10 , H12 /H11 ∼ = K21 /K11 , H13 /H12 ∼ = K22 /K12 . (Notice the pattern; in each isomorphism, the ﬁrst subscript in the numerator is increased by 1 and the second subscript is decreased by 1 in going from left to right. The subscripts in the denominator are unchanged.) The factor groups of the second series are a reordering of the factor groups of the ﬁrst series. ♣ The hard work is now accomplished, and we have everything we need to prove the main result. 5.6.6 Jordan-H¨ older Theorem If G has a composition series S (in particular if G is ﬁnite), then any subnormal series R without repetition can be reﬁned to a composition series. Furthermore, any two composition series for G are equivalent. Proof. By (5.6.5), R and S have equivalent reﬁnements. Remove any repetitions from the reﬁnements to produce equivalent reﬁnements R0 and S0 without repetitions. But a composition series has no proper reﬁnements, hence S0 = S, proving the ﬁrst assertion. If R is also a composition series, then R0 = R as well, and R is equivalent to S. ♣ Problems For Section 5.6 1. Show that if G has a composition series, so does every normal subgroup of G. 2. Give an example of a group that has no composition series.

page 15 of Chapter 5

3. Give an example of two nonisomorphic groups with the same composition factors, up to rearrangement. Problems 4-9 will prove that the alternating group An is simple for all n ≥ 5. (A1 and A2 are trivial and hence not simple; A4 is not simple by the example given in (5.6.1); A3 is cyclic of order 3 and is therefore simple.) In these problems, N stands for a normal subgroup of An . 4. Show that if n ≥ 3, then An is generated by 3-cycles. 5. Show that if N contains a 3-cycle, then it contains all 3-cycles, so that N = An . 6. From now on, assume that N is a proper normal subgroup of An . Show that no permutation in N contains a cycle of length 4 or more. 7. Show that no permutation in N contains the product of two disjoint 3-cycles. Thus in view of Problems 4,5 and 6, every member of N is the product of an even number of disjoint transpositions. 8. In Problem 7, show that the number of transpositions is at least 4. 9. Finally, show that the assumption that N contains a product of 4 or more disjoint transpositions leads to a contradiction, proving that N = 1, so that An is simple. It follows that a composition series for Sn is 1 ) An ) Sn . 10. A chief series is a normal series without repetition that cannot be properly reﬁned to another normal series. Show that if G has a chief series, then any normal series without repetition can be reﬁned to a chief series. Furthermore, any two chief series of a given group are equivalent. 11. In a composition series, the factor groups Gi+1 /Gi are required to be simple. What is the analogous condition for a chief series? 5.7 Solvable And Nilpotent Groups Solvable groups are so named because of their connection with solvability of polynomial equations, a subject to be explored in the next chapter. To get started, we need a property of subgroups that is stronger than normality. 5.7.1 Deﬁnitions and Comments A subgroup H of the group G is characteristic (in G) if for each automorphism f of G, f (H) = H. Thus f restricted to H is an automorphism of H. Consequently, if H is characteristic in G, then it is normal in G. If follows from the deﬁnition that if H is characteristic in K and K is characteristic in G, then H is characteristic in G. Another useful result is the following. (1) If H is characteristic in K and K is normal in G, then H is normal in G. To see this, observe that any inner automorphism of G maps K to itself, so restricts to an automorphism (not necessarily inner) of K. Further restriction to H results in an automorphism of H, and the result follows. 5.7.2 More Deﬁnitions and Comments The commutator subgroup G of a group G is the subgroup generated by all commutators [x, y] = xyx−1 y −1 . (Since [x, y]−1 = [y, x], G consists of all ﬁnite products of commutators.) Here are some basic properties. (2) G is characteristic in G. This follows because any automorphism f maps a commutator to a commutator: f [x, y] = [f (x), f (y)]. (3) G is abelian if and only if G is trivial. This holds because [x, y] = 1 iﬀ xy = yx. (4) G/G is abelian. Thus forming the quotient of G by G , sometimes called modding out by G , in a sense “abelianizes” the group.

page 16 of Chapter 5

For G xG y = G yG x iﬀ G xy = G yx iﬀ xy(yx)−1 ∈ G iﬀ xyx−1 y −1 ∈ G , and this holds for all x and y by deﬁnition of G . (5) If N G, then G/N is abelian if and only if G ≤ N . The proof of (4) with G replaced by N shows that G/N is abelian iﬀ all commutators belong to N , that is, iﬀ G ≤ N . The process of taking commutators can be iterated: G(0) = G, G(1) = G , G(2) = (G ) , and in general,

G(i+1) = (G(i) ) , i = 0, 1, 2, . . . .

Since G(i+1) is characteristic in G(i) , an induction argument shows that each G(i) is characteristic, hence normal, in G. The group G is said to be solvable if G(r) = 1 for some r. We then have a normal series 1 = G(r) G(r−1) · · · G(0) = G called the derived series of G. Every abelian group is solvable, by (3). Note that a group that is both simple and solvable must be cyclic of prime order. For the normal subgroup G must be trivial; if it were G, then the derived series would never reach 1. By (3), G is abelian, and by (5.5.1), G must be cyclic of prime order. A nonabelian simple group G (such as An , n ≥ 5) cannot be solvable. For if G is nonabelian, then G is not trivial. Thus G = G, and as in the previous paragraph, the derived series will not reach 1. There are several equivalent ways to describe solvability. 5.7.3 Proposition The following conditions are equivalent. (i) G is solvable. (ii) G has a normal series with abelian factors. (iii) G has a subnormal series with abelian factors. Proof. Since (i) implies (ii) by (4) and (ii) implies (iii) by deﬁnition of normal and subnormal series, the only problem is (iii) implies (i). Suppose G has a subnormal series 1 = Gr Gr−1 · · · G1 G0 = G with abelian factors. Since G/G1 is abelian, we have G ≤ G1 by (5), and an induction argument then shows that G(i) ≤ Gi for all i. [The inductive step is G(i+1) = (G(i) ) ≤ Gi ≤ Gi+1 since Gi /Gi+1 is abelian.] Thus G(r) ≤ Gr = 1. ♣ The next result gives some very useful properties of solvable groups. 5.7.4 Proposition Subgroups and quotients of a solvable group are solvable. Conversely, if N is normal subgroup of G and both N and G/N are solvable, then G is solvable. Proof. If H is a subgroup of the solvable group G, then H is solvable because H (i) ≤ G(i) for all i. If N is a normal subgroup of the solvable group G, observe that commutators of G/N look like xyx−1 y −1 N , so (G/N ) = G N/N . (Not G /N , since N is not necessarily a subgroup of G .) Inductively, (G/N )(i) = G(i) N/N

page 17 of Chapter 5

and since N/N is trivial, G/N is solvable. Conversely, suppose that we have a subnormal series from N0 = 1 to Nr = N , and a subnormal series from G0 /N = 1 (i.e., G0 = N ) to Gs /N = G/N (i.e., Gs = G) with abelian factors in both cases. Then we splice the series of Ni ’s to the series of Gi ’s. The latter series is subnormal by the correspondence theorem, and the factors remain abelian by the third isomorphism theorem. ♣ 5.7.5 Corollary If G has a composition series, in particular if G is ﬁnite, then G is solvable if and only if the composition factors of G are cyclic of prime order. Proof. Let Gi+1 /Gi be a composition factor of the solvable group G. By (5.7.4), Gi+1 is solvable, and again by (5.7.4), Gi+1 /Gi is solvable. But a composition factor must be a simple group, so Gi+1 /Gi is cyclic of prime order, as observed in (5.7.2). Conversely, if the composition factors of G are cyclic of prime order, then the composition series is a subnormal series with abelian factors. ♣ Nilpotent groups arise from a diﬀerent type of normal series. We will get at this idea indirectly, and give an abbreviated treatment. 5.7.6 Proposition If G is a ﬁnite group, the following conditions are equivalent, and deﬁne a nilpotent group. [Nilpotence of an arbitrary group will be deﬁned in (5.7.8).] (a) G is the direct product of its Sylow subgroups. (b) Every Sylow subgroup of G is normal. Proof. (a) implies (b): By (1.5.3), the factors of a direct product are normal subgroups. (b) implies (a): By (5.5.4), there is a unique Sylow pi -subgroup Hi for each prime divisor pi of |G|, i = 1, . . . , k. By successive application of (5.2.4), we have |H1 · · · Hk | = |H1 | · · · |Hk |, which is |G| by deﬁnition of Sylow p-subgroup. Since all sets are ﬁnite, G = H1 · · · Hk . Furthermore, each Hi ∩ j=i Hj is trivial, because the orders of the Hi are distinct primes. By (1.5.4), G is the direct product of the Hi . ♣ 5.7.7 Corollary Every ﬁnite abelian group and every ﬁnite p-group is nilpotent. Proof. The analysis in (4.6.4) shows that a ﬁnite abelian group is the direct product of its Sylow subgroups, and is therefore nilpotent. If P is a ﬁnite p-group, then P has only one Sylow subgroup, P itself, so the conditions of (5.7.6) are automatically satisﬁed. ♣ We now connect this discussion with normal series. Suppose that we are trying to build a normal series for the group G, starting with G0 = 1. We take G1 to be Z(G), the center of G; we have G1 G by (5.1.3), Example 3. We deﬁne G2 by the correspondence theorem: G2 /G1 = Z(G/G1 ) and since Z(G/G1 ) G/G1 , we have G2 G. In general, we take Gi /Gi−1 = Z(G/Gi−1 ), and by induction we have Gi G. The diﬃculty is that there is no guarantee that Gi will ever reach G. However, we will succeed if G is a ﬁnite p-group. The key point is that a nontrivial ﬁnite p-group has a nontrivial center, by (5.5.3). Thus by induction, Gi /Gi−1 is nontrivial for every i, so Gi−1 < Gi . Since G is ﬁnite, it must eventually be reached. 5.7.8 Deﬁnitions and Comments A central series for G is a normal series 1 = G0 G1 · · · Gr = G such that Gi /Gi−1 ⊆ Z(G/Gi−1 ) for every i = 1, . . . , r. (The series just discussed is a special case called the upper central series.) An arbitrary group G is said to be nilpotent if it has a central series. Thus a ﬁnite p-group is nilpotent, and in particular, every Sylow p-subgroup is nilpotent. Now a direct product of a ﬁnite number of nilpotent groups th is nilpotent. (If Gij is the ith term of a central Hj , with Gij = G if series of the j factor the series has already terminated at G, then j Gij will be the ith term of a central series

page 18 of Chapter 5

for j Hj .) Thus a ﬁnite group that satisﬁes the conditions of (5.7.6) has a central series. Conversely, it can be shown that a ﬁnite group that has a central series satisﬁes (5.7.6), so the two deﬁnitions of nilpotence agree for ﬁnite groups. Note that a nilpotent group is solvable. For if Gi /Gi−1 ⊆ Z(G/Gi−1 ), then the elements of Gi /Gi−1 commute with each other since they commute with everything in G/Gi−1 ; thus Gi /Gi−1 is abelian. Consequently, a ﬁnite p-group is solvable. Problems For Section 5.7 1. Give an example of a nonabelian solvable group. 2. Show that a solvable group that has a composition series must be ﬁnite. 3. Prove directly (without making use of nilpotence) that a ﬁnite p-group is solvable. 4. Give an example of a solvable group that is not nilpotent. 5. Show that if n ≥ 5, then Sn is not solvable. 6. If P is a ﬁnite simple p-group, show that P has order p. 7. Let P be a nontrivial ﬁnite p-group. Show that P has a normal subgroup N whose index [P : N ] is p. 8. Let G be a ﬁnite group of order pr m, where r is a positive integer and p does not divide m. Show that for any k = 1, 2, . . . , r, G has a subgroup of order pk . 9. Give an example of a group G with a normal subgroup N such that N and G/N are abelian, but G is not abelian. Thus (5.7.4) does not hold for abelian groups. 10. If G is a solvable group, its derived length, dl(G), is the smallest nonnegative integer r such that G(r) = 1. If N is a normal subgroup of the solvable group G, what can be said about the relation between dl(G), dl(N ) and dl(G/N )? 5.8 Generators And Relations In (1.2.4) we gave an informal description of the dihedral group via generators and relations, and now we try to make the ideas more precise. 5.8.1 Deﬁnitions and Comments The free group G on the set S (or the free group with basis S) consists of all words on S, that is, all ﬁnite sequences x1 · · · xn , n = 0, 1, . . ., where each xi is either an element of S or the inverse of an element of S. We regard the case n = 0 as the empty word λ. The group operation is concatenation, subject to the constraint that if s and s−1 occur in succession, they can be cancelled. The empty word is the identity, and inverses are calculated in the only reasonable way, for example, (stu)−1 = u−1 t−1 s−1 . We say that G is free on S. Now suppose that G is free on S, and we attempt to construct a homomorphism f from G to an arbitrary group H. The key point is that f is completely determined by its values on S. If f (s1 ) = a, f (s2 ) = b, f (s3 ) = c, then −1 f (s1 s−1 f (s3 ) = ab−1 c. 2 s3 ) = f (s1 )f (s2 )

Here is the formal statement, followed by an informal proof. 5.8.2 Theorem If G is free on S and g is an arbitrary function from S to a group H, then there is a unique homomorphism f : G → H such that f = g on S. Proof. The above discussion is a nice illustration of a concrete example with all the features of the general case. The analysis shows both existence and uniqueness of f . A formal proof must show that all aspects of the general case are covered. For example, if u = s1 s−1 2 s3 −1 −1 and v = s1 s−1 s s s , then f (u) = f (v), so that cancellation of s s causes no diﬃculty. 4 3 4 2 4 4 Speciﬁc calculations of this type are rather convincing, and we will not pursue the formal details. (See, for example, Rotman, An Introduction to the Theory of Groups, p.343-345.) ♣ 5.8.3 Corollary Any group H is a homomorphic image of a free group.

page 19 of Chapter 5

Proof. Let S be a set of generators for H (if necessary, take S = H), and let G be free on S. Deﬁne g(s) = s for all s ∈ S. If f is the unique extension of g to G, then since S generates H, f is an epimorphism. ♣ Returning to (1.2.4), we described a group H using generators R and F , and relations Rn = I, F 2 = I, RF = F R−1 . The last relation is equivalent to RF RF = I, since F 2 = I. The words Rn , F 2 and RF RF are called relators, and the speciﬁcation of generators and relations is called a presentation. We use the notation H =< R, F | Rn , F 2 , RF RF > or the long form

H =< R, F | Rn = I, F 2 = I, RF = F R−1 > .

We must say precisely what it means to deﬁne a group by generators and relations, and show that the above presentation yields a group isomorphic to the dihedral group D2n . We start with the free group on {R, F } and set all relators equal to the identity. It is natural to mod out by the subgroup generated by the relators, but there is a technical diﬃculty; this subgroup is not necessarily normal. 5.8.4 Deﬁnition Let G be free on the set S, and let K be a subset of G. We deﬁne the group < S | K > as G/K, where K is the smallest normal subgroup of G containing K. Unfortunately, it is a theorem of mathematical logic that there is no algorithm which when given a presentation, will ﬁnd the order of the group. In fact, there is no algorithm to determine whether a given word of < S | K > coincides with the identity. Logicians say that the word problem for groups is unsolvable. But although there is no general solution, there are speciﬁc cases that can be analyzed, and the following result is very helpful. 5.8.5 Von Dyck’s Theorem Let H =< S | K > be a presentation, and let L be a group that is generated by the words in S. If L satisﬁes all the relations of K, then there is an epimorphism α : H → L. Consequently, |H| ≥ |L|. Proof. Let G be free on S, and let i be the identity map from S, regarded as a subset of G, to S, regarded as a subset of L. By (5.8.2), i has a unique extension to a homomorphism f of G into L, and in fact f is an epimorphism because S generates L. Now f maps any word of G to the same word in L, and since L satisﬁes all the relations, we have K ⊆ ker f . But the kernel of f is a normal subgroup of G, hence K ⊆ ker f . The factor theorem provides an epimorphism α : G/K → L. ♣ 5.8.6 Justifying a presentation If L is a ﬁnite group generated by the words of S, then in practice, the crucial step in identifying L with H =< S | K > is a proof that |H| ≤ |L|. If we can accomplish this, then by (5.8.5), |H| = |L|. In this case, α is a surjective map of ﬁnite sets of the same size, so α is injective as well, hence is an isomorphism. For the dihedral group we have H =< F, R | Rn , F 2 , RF RF > and L = D2n . In (1.2.4) we showed that each word of H can be expressed as Ri F j with 0 ≤ i ≤ n − 1 and 0 ≤ j ≤ 1. Therefore |H| ≤ 2n = |D2n | = |L|. Thus the presentation H is a legitimate description of the dihedral group. Problems For Section 5.8 1. Show that a presentation of the cyclic group of order n is < a | an >. 2. Show that the quaternion group (see (2.1.3, Example 4)) has a presentation < a, b | a4 = 1, b2 = a2 , ab = ba−1 >. 3. Show that H =< a, b | a3 = 1, b2 = 1, ba = a−1 b > is a presentation of S3 . 4. Is the presentation of a group unique? In Problems 5-11, we examine a diﬀerent way of assembling a group from subgroups, which generalizes the notion of a direct product. Let N be a normal subgroup of G, and H

page 20 of Chapter 5

an arbitrary subgroup. We say that G is the semidirect product of N by H if G = N H and N ∩ H = 1. (If H G, we have the direct product.) For notational convenience, the letter n, possibly with subscripts, will always indicate a member of N , and similarly h will always belong to H. In Problems 5 and 6, we assume that G is the semidirect product of N by H. 5. If n1 h1 = n2 h2 , show that n1 = n2 and h1 = h2 . 6. If i : N → G is inclusion and π : G → H is projection (π(nh) = h), then the sequence 1

→

N

i → G

π →

H

→ 1

is exact. Show that the sequence splits on the right, i.e., there is a homomorphism ψ : H → G such that π ◦ ψ = 1. 7. Conversely, suppose that the above exact sequence splits on the right. Since ψ is injective, we can regard H (and N as well) as subgroups of G, with ψ and i as inclusion maps. Show that G is the semidirect product of N by H. 8. Let N and H be arbitrary groups, and let f be a homomorphism of H into Aut N , the group of automorphisms of N . Deﬁne a multiplication on G = N × H by (n1 , h1 )(n2 , h2 ) = (n1 f (h1 )(n2 ), h1 h2 ). [f (h1 )(n2 ) is the value of the automorphism f (h1 ) at the element n2 .] A lengthy but straightforward calculation shows that G is a group with identity (1,1) and inverses given by (n, h)−1 = (f (h−1 )(n−1 ), h−1 ). Show that G is the semidirect product of N × {1} and {1} × H. 9. Show that every semidirect product arises from the construction of Problem 8. 10. Show by example that it is possible for a short exact sequence of groups to split on the right but not on the left. [If h : G → N is a left-splitting map in the exact sequence of Problem 6, then h and π can be used to identify G with the direct product of N and H. Thus a left-splitting implies a right-splitting, but, unlike the result for modules in (4.7.4), not conversely.] 11. Give an example of a short exact sequence of groups that does not split on the right. 12. (The Frattini argument, frequently useful in a further study of group theory.) Let N be a normal subgroup of the ﬁnite group G, and let P be a Sylow p-subgroup of N . If NG (P ) is the normalizer of P in G, show that G = NG (P )N (= N NG (P ) by (1.4.3)).[If g ∈ G, look at the relation between P and gP g −1 .] 13. Let N = {1, a, a2 , . . . , an−1 } be a cyclic group of order n, and let H = {1, b} be a cyclic group of order 2. Deﬁne f : H → Aut N by taking f (b) to be the automorphism that sends a to a−1 . Show that the dihedral group D2n is the semidirect product of N by H. (See Problems 8 and 9 for the construction of the semidirect product.) 14. In Problem 13, replace N by an inﬁnite cyclic group {. . . , a−2 , a−1 , 1, a, a2 , . . .}. Give a presentation of the semidirect product of N by H. This group is called the inﬁnite dihedral group D∞ . Concluding Remarks Suppose that the ﬁnite group G has a composition series 1 = G0 ) G1 ) · · · ) Gr = G. If Hi = Gi /Gi−1 , then we say that Gi is an extension of Gi−1 by Hi in the sense that Gi−1 Gi and Gi /Gi−1 ∼ = Hi . If we were able to solve the extension problem (ﬁnd all possible extensions of Gi−1 by Hi ) and we had a catalog of all ﬁnite simple groups, then we could build a catalog of all ﬁnite groups. This sharpens the statement made in (5.6.1) about the importance of simple groups.

page 1 of Chapter 6

CHAPTER 6

GALOIS THEORY

6.1 Fixed Fields and Galois Groups Galois theory is based on a remarkable correspondence between subgroups of the Galois group of an extension E/F and intermediate fields between E and F . In this section we will set up the machinery for the fundamental theorem. [A remark on notation: Throughout the chapter, the composition τ ◦ σ of two automorphisms will be written as a product τ σ.] 6.1.1 Definitions and Comments Let G =Gal(E/F ) be the Galois group of the extension E/F . If H is a subgroup of G, the fixed field of H is the set of elements fixed by every automorphism in H, that is, F (H) = {x ∈ E : σ(x) = x for every σ ∈ H}. If K is an intermediate field, that is, F ≤ K ≤ E, define G(K) = Gal(E/K) = {σ ∈ G : σ(x) = x for every x ∈ K}. I like the term “fixing group of K” for G(K), since G(K) is the group of automorphisms of E that leave K fixed. Galois theory is about the relation between fixed fields and fixing groups. In particular, the next result suggests that the smallest subfield F corresponds to the largest subgroup G. 6.1.2 Proposition Let E/F be a finite Galois extension with Galois group G =Gal(E/F ). Then (i) The fixed field of G is F ; (ii) If H is a proper subgroup of G, then the fixed field of H properly contains F . Proof. (i) Let F0 be the fixed field of G. If σ is an F -automorphism of E, then by definition of F0 , σ fixes everything in F0 . Thus the F -automorphisms of G coincide with the F0-automorphisms of G. Now by (3.4.7) and (3.5.8), E/F0 is Galois. By (3.5.9), the size of the Galois group of a finite Galois extension is the degree of the extension. Thus [E : F ] = [E : F0 ], so by (3.1.9), F = F0. (ii) Suppose that F = F (H). By the theorem of the primitive element (3.5.12), we have E = F (α) for some α ∈ E. Define a polynomial f (X) ∈ E[X] by f(X) =

Y

(X − σ(α)).

σ∈H

If τ is any automorphism in H, then we may apply τ to f (that is, to the coefficients of f; we discussed this idea in the proof of (3.5.2)). The result is (τ f )(X) =

Y

(X − (τ σ)(α)).

σ∈H

But as σ ranges over all of H, so does τ σ, and consequently τ f = f . Thus each coefficient of f is fixed by H, so f ∈ F [X]. Now α is a root of f , since X − σ(α) is 0 when X = α and σ is the identity. We can say two things about the degree of f : (1) By definition of f , deg f = |H| < |G| = [E : F ], and, since f is a multiple of the minimal polynomial of α over F , (2) deg f ≥ [F (α) : F ] = [E : F ], and we have a contradiction. ♣ There is a converse to the first part of (6.1.2). 6.1.3 Proposition Let E/F be a finite extension with Galois group G. If the fixed field of G is F , then E/F is Galois. Proof. Let G = {σ1 , . . . , σn }, where σ1 is the identity. To show that E/F is normal, we consider an irreducible polynomial f ∈ F [X] with a root α ∈ E. Apply each automorphism in G to α, and suppose that

page 2 of Chapter 6

there are r distinct images α = α1 = σ1 (α), α2 = σ2 (α), . . . , αr = σr (α). If σ is any member of G, then σ will map each αi to some αj , and since σ is an injective map of the finite set {α1 , . . . , αr } to itself, it is surjective as well. To put it simply, σ permutes the αi . Now we examine what σ does to the elementary symmetric functions of the αi , which are given by e1 =

n X i=1

αi , e2 =

X

αi αj , e3 =

i<j

er =

X

αi αj αk , . . . ,

i<j
αi .

i=1

Since σ permutes the αi , it follows that σ(ei ) = ei for all i. Thus the ei belong to the fixed field of G, which is F by hypothesis. Now we form a monic polynomial whose roots are the αi : g(X) = (X − α1 ) · · · (X − αr ) = X r − e1 X r−1 + e2 X r−2 − · · · + (−1)r er . Since the ei belong to F , g ∈ F [X], and since the αi are in E, g splits over E. We claim that g is the minimal polynomial of α over F . To see this, let h(X ) = b0 + b1 X + · · · + bm X m be any polynomial in F [X] having α as a root. Applying σi to the equation b0 + b1α + · · · bm αm = 0 we have b0 + b1 αi + · · · bm αm i = 0, so that each αi is a root of h, hence g divides h and therefore g =min(α, F ). But our original polynomial f ∈ F [X] is irreducible and has α as a root, so it must be a constant multiple of g. Consequently, f splits over E, proving that E/F is normal. Since the αi , i = 1, . . . r, are distinct, g has no repeated roots. Thus α is separable over F , which shows that the extension E/F is separable. ♣ It is profitable to examine elementary symmetric functions in more detail. 6.1.4 Theorem Let f be a symmetric polynomial in the n variables X1 , . . . , Xn . [This means that if σ is any permutation in Sn and we replace Xi by Xσ(i) for i = 1, . . . , n, then f is unchanged.] If e1 , . . . , en are the elementary symmetric functions of the Xi , then f can be expressed as a polynomial in the ei . Proof. We give an algorithm. The polynomial f is a linear combination of monomials of the form X1r1 · · · Xnrn , and we order the monomials lexicographically: X1r1 · · · Xnrn > X1s1 · · · Xnsn iff the first disagreement between ri and si results in ri > si . Since f is symmetric, all terms generated by applying a permutation σ ∈ Sn to the subscripts of X1r1 · · · Xnrn will also contribute to f . The idea is to cancel the leading terms (those associated with the monomial that is first in the ordering) by subtracting an expression of the form et11 et22 · · · etnn = (X1 + · · · + Xn )t1 · · · (X1 · · · Xn )tn which has leading term X1t1 (X1 X2 )t2 (X1 X2 X3 )t3 · · · (X1 · · · Xn )tn = X1t1 +···+tn X2t2 +···+tn · · · Xntn . This will be possible if we choose t1 = r1 − r2 , t2 = r2 − r3, . . . , tn−1 = rn−1 − rn , tn = rn . After subtraction, the resulting polynomial has a leading term that is below X1r1 · · · Xnrn in the lexicographical ordering. We can then repeat the procedure, which must terminate in a finite number of steps. ♣ 6.1.5 Corollary If g is a polynomial in F [X ] and f(α1 , . . . , αn ) is any symmetric polynomial in the roots α1 , . . . , αn of g, then f ∈ F [X].

page 3 of Chapter 6

Proof. We may assume without loss of generality that g is monic. Then in a splitting field of g we have g(X) = (X − α1 ) · · · (X − αn ) = X n − e1 X n−1 + · · · + (−1)n en . By (6.1.4), f is a polynomial in the ei , and since the ei are simply ± the coefficients of g, the coefficients of f are in F . ♣ 6.1.6 Dedekind’s Lemma The result that the size of the Galois group of a finite Galois extension is the degree of the extension can be proved via Dedekind’s lemma, which is of interest in its own right. Let G be a group and E a field. A character from G to E is a homomorphism from G to the multiplicative group E ∗ of nonzero elements of E. In particular, an automorphism of E defines a character with G = E ∗ , as does a monomorphism of E into a field L. Dedekind’s lemma states that if σ1 , . . . , σn are distinct characters from G to E, then the σi are linearly independent over E. The proof is given in Problems 3 and 4. Problems For Section 6.1 1. Express X12 X2 X3 + X1X22X3 + X1 X2 X32 in terms of elementary symmetric functions. 2. Repeat Problem 1 forX12X2 + X12 X3 + X1 X22 + X1 X32 + X22 X3 + X2X32 + 4X1 X2 X3 . 3. To begin the proof of Dedekind’s lemma, suppose that the σi are linearly dependent. By renumbering the σi if necessary, we have a1 σ1 + · · · ar σr = 0 where all ai are nonzero and r is as small as possible. Show that for every h and g ∈ G, we have r X

ai σ1 (h)σi (g) = 0

(1)

ai σi (h)σi (g) = 0.

(2)

i=1

and

r X i=1

[Equations (1) and (2) are not the same; in (1) we have σ1 (h), not σi (h).] 4. Continuing Problem 3, subtract (2) from (1) to get r X

ai (σ1(h) − σi (h))σi (g) = 0.

(3)

i=1

With g arbitrary, reach a contradiction by an appropriate choice of h. √ 5. If G is the Galois group of Q( 3 2) over Q, what is the fixed field of G? 6. Find the Galois group of C/R. 7. Find the fixed field of the Galois group of Problem 6. 6.2 The Fundamental Theorem With the preliminaries now taken care of, we can proceed directly to the main result. 6.2.1 Fundamental Theorem of Galois Theory Let E/F be a finite Galois extension with Galois group G. If H is a subgroup of G, let F (H) be the fixed field of H, and if K is an intermediate field, let G(K) be Gal(E/K), the fixing group of K (see (6.1.1)). (1) F is a bijective map from subgroups to intermediate fields, with inverse G. Both maps are inclusionreversing, that is, if H1 ≤ H2 then F (H1 ) ≥ F(H2 ), and if K1 ≤ K2 , then G(K1 ) ≥ G(K2 ). (2) Suppose that the intermediate field K corresponds to the subgroup H under the Galois correspondence. Then (a) E/K is always normal (hence Galois); (b) K/F is normal if and only if H is a normal subgroup of G, and in this case, (c) the Galois group of K/F is isomorphic to the quotient group G/H. Moreover, whether or not K/F is

page 4 of Chapter 6

normal, (d) [K : F ] = [G : H] and [E : K] = |H|. (3) If the intermediate field K corresponds to the subgroup H and σ is any automorphism in G, then the field σK = {σ(x) : x ∈ K} corresponds to the conjugate subgroup σHσ −1 . For this reason, σK is called a conjugate subfield of K. The following diagram may aid the understanding. E | K | F

G | H | 1

As we travel up the left side from smaller to larger fields, we move down the right side from larger to smaller groups. A statement about K/F , an extension at the bottom of the left side, corresponds to a statement about G/H, located at the top of the right side. Similarly, a statement about E/K corresponds to a statement about H/1 = H. Proof. (1) First, consider the composite mapping H → F (H) → GF (H). If σ ∈ H then σ fixes F (H) by definition of fixed field, and therefore σ ∈ GF(H) =Gal(E/F (H)). Thus H ⊆ GF(H). If the inclusion is proper, then by (6.1.2) part (ii) with F replaced by F (H), we have F (H) > F (H), a contradiction. [Note that E/K is a Galois extension for any intermediate field K, by (3.4.7) and (3.5.8).] Thus GF (H) = H. Now consider the mapping K → G(K) → F G(K) = FGal(E/K). By (6.1.2) part (i) with F replaced by K, we have F G(K) = K. Since both F and G are inclusion-reversing by definition, the proof of (1) is complete. (3) The fixed field of σH σ −1 is the set of all x ∈ E such that στσ −1(x) = x for every τ ∈ H. Thus F (σHσ −1 ) = {x ∈ E : σ −1 (x) ∈ F (H)} = σ(F (H)). (2a) This was observed in the proof of (1). (2b) If σ is an F -monomorphism of K into E, then by (3.5.2) and (3.5.6), σ extends to an F -monomorphism of E into itself, in other words (see (3.5.6)), an F -automorphism of E. Thus each such σ is the restriction to K of a member of G. Conversely, the restriction of an automorphism in G to K is an F -monomorphism of K into E. By (3.5.5) and (3.5.6), K/F is normal iff for every σ ∈ G we have σ(K) = K. But by (3), σ(K) corresponds to σHσ −1 and K to H. Thus K/F is normal iff σHσ −1 = H for every σ ∈ G, i.e., K E G. (2c) Consider the homomorphism of G =Gal(E/F ) to Gal(K/F ) given by σ → σ |K . The map is surjective by the argument just given in the proof of (2b). The kernel is the set of all automorphisms in G that restrict to the identity on K, that is, Gal(E/K) = H. The result follows from the first isomorphism theorem. (2d) By (3.1.9), [E : F ] = [E : K][K : F ]. The term on the left is |G| by (3.5.9), and the first term on the right is |Gal(E/K)| by (2a), and this in turn is |H| since H = G(K). Thus |G| = |H|[K : F ], and the result follows from Lagrange’s theorem. [If K/F is normal, the proof is slightly faster. The first statement follows from (2c). To prove the second, note that by (3.1.9) and (3.5.9), [E : K] =

[E : F ] |G| = = |H|.] ♣ [K : F ] |G/H|

The next result is reminiscent of the second isomorphism theorem, and is best visualized via the diamond diagram of Figure 6.2.1. In the diagram, EK is the composite of the two fields E and K, that is, the smallest field containing both E and K.

page 5 of Chapter 6

6.2.2 Theorem Let E/F be a finite Galois extension and K/F an arbitrary extension. Assume that E and K are both contained in a common field, so that it is sensible to consider the composite EK. Then (1) EK/K is a finite Galois extension; (2) Gal(EK/K) is embedded in Gal(E/F ), where the embedding is accomplished by restricting automorphisms in Gal(EK/K) to E; (3) The embedding is an isomorphism if and only if E ∩ K = F . Proof. (1) By the theorem of the primitive element (3.5.12), we have E = F [α] for some α ∈ E, so EK = KF [α] = K[α]. The extension K[α]/K is finite because α is algebraic over F , hence over K. Since α, regarded as an element of EK, is separable over F and hence over K, it follows that EK/K is separable. [To avoid breaking the main line of thought, this result will be developed in the exercises (see Problems 1 and 2).] Now let f be the minimal polynomial of α over F , and g the minimal polynomial of α over K. Since f ∈ K[X] and f (α) = 0, we have g|f , and the roots of g must belong to E ⊆ EK = K[α] because E/F is normal. Therefore K[α] is a splitting field for g over K, so by (3.5.7), K[α]/K is normal. (2) If σ is an automorphism in Gal(EK/K), restrict σ to E, thus defining a homomorphism from Gal(EK/K) to Gal(E/F ). (Note that σ|E is an automorphism of E because E/F is normal.) Now σ fixes K, and if σ belongs to the kernel of the homomorphism, then σ also fixes E, so σ fixes EK = K[α]. Thus σ is the identity, and the kernel is trivial, proving that the homomorphism is actually an embedding. (3) The embedding of (2) maps Gal(EK/K) to a subgroup H of Gal(E/F ), and we will find the fixed field of H. By (6.1.2), the fixed field of Gal(EK/K) is K, and since the embedding just restricts automomorphisms to E, the fixed field of H must be E ∩ K. By the fundamental theorem, H =Gal(E/(E ∩ K)). Thus H = Gal(E/F ) iff Gal(E/(E ∩ K)) = Gal(E/F ), and by applying the fixed field operator F, we see that this happens if and only if E ∩ K = F . ♣ Problems For Section 6.2 1. Let E = F (α1 , . . . , αn ), where each αi is algebraic and separable over F . We are going to show that E is separable over F . Without loss of generality, we can assume that the characteristic of F is a prime p, and since F/F is separable, the result holds for n = 0. To carry out the inductive step, let Ei = F (α1 , . . . , αi ), p so that Ei+1 = Ei (αi+1 ). Show that Ei+1 = Ei (Ei+1 ). (See Section 3.4, Problems 4-8, for the notation.) 2. Continuing Problem 1, show that E is separable over F . 3. Let E = F (α1 , . . . , αn ), where each αi is algebraic over F . If for each i = 1, . . . , n, all the conjugates of αi (the roots of the minimal polynomial of αi over F ) belong to E, show that E/F is normal. 4. Suppose that F = K0 ≤ K1 ≤ · · · ≤ Kn = E, where E/F is a finite Galois extension, and that the intermediate field Ki corresponds to the subgroup Hi under the Galois correspondence. Show that Ki /Ki−1 is normal (hence Galois) if and only if Hi E Hi−1, and in this case, Gal(Ki /Ki−1 ) is isomorphic to Hi−1 /Hi . 5. Let E and K be extensions of F , and assume that the composite EK is defined. If A is any set of generators for K over F (for example, A = K), show that EK = E(A), the field formed from E by adjoining the elements of A.

page 6 of Chapter 6

6. Let E/F be a finite Galois extension with Galois group G, and let E 0 /F 0 be a finite Galois extension with Galois group G0 . If τ is an isomorphism of E and E 0 with τ (F ) = F 0 , we expect intuitively that G ∼ = G0 . Prove this formally. 7. Let K/F be a finite separable extension. Although K need not be a normal extension of F , we can form the normal closure N of K over F , as in (3.5.11). Then N/F is a Galois extension (see Problem 8 of Section 6.3); let G be its Galois group. Let H = Gal (N/K), so that the fixed field of H is K. If H 0 is a normal subgroup of G that is contained in H, show that the fixed field of H 0 is N . 8. Continuing Problem 7, show that H 0 is trivial, and conclude that \ gHg −1 = {1} g∈G

where 1 is the identity automorphism. 6.3 Computing a Galois Group Directly 6.3.1 Definitions and Comments Suppose that E is a splitting field of the separable polynomial f over F . The Galois group of f is the Galois group of the extension E/F . (The extension is indeed Galois; see Problem 8.) Given f , how can we determine its Galois group? It is not so easy, but later we will develop a systematic approach for polynomials of degree 4 or less. Some cases can be handled directly, and in this section we look at a typical situation. A useful observation is that the Galois group G of a finite Galois extension E/F acts transitively on the roots of any irreducible polynomial h ∈ F [X] (assuming that one, hence every, root of h belongs to E). [Each σ ∈ G permutes the roots by (3.5.1). If α and β are roots of h, then by (3.2.3) there is an F -isomorphism of F (α) and F (β) carrying α to β. This isomorphism can be extended to an F -automorphism of E by (3.5.2), (3.5.5) and (3.5.6).] 6.3.2 Example Let d be a √ positive integer that is not a perfect√cube, and let θ be the positive cube root of d. Let ω = ei2π/3 = − 12 + i 12 3, so that ω 2 = e−i2π/3 = − 12 − i 12 3 = −(1 + ω). The minimal polynomial of θ over the rationals Q is f (X) = X 3 − d, because if f were reducible then it would have a linear factor and d would be a perfect cube. The minimal polynomial of ω over Q is g(X ) = X 2 + X + 1. (If g were reducible, it would have a rational (hence real) root, so the discriminant would be nonnegative, a contradiction.) We will compute the Galois group G of the polynomial f(X)g(X ), which is the Galois group of E = Q(θ, ω) over Q. If the degree of E/Q is the product of the degrees of f and g, we will be able to make progress. We have [Q(θ) : Q] = 3 and, since ω, a complex number, does not belong to Q(θ), we have [Q(θ, ω) : Q(θ)] = 2. Thus [Q(θ, ω) : Q] = 6. But the degree of a finite Galois extension is the size of the Galois group by (3.5.9), so G has exactly 6 automorphisms. Now any σ ∈ G must take θ to one of its conjugates, namely θ, ωθ or ω 2 θ. Moreover, σ must take ω to a conjugate, namely ω or ω 2 . Since σ is determined by its action on θ and ω, we have found all 6 members of G. The results can be displayed as follows. 1 : θ → θ, ω → ω, order = 1 τ : θ → θ, ω → ω 2 , order = 2 σ : θ → ωθ, ω → ω, order = 3 στ : θ → ωθ, ω → ω 2 , order = 2 σ 2 : θ → ω 2 θ, ω → ω, order = 3 τ σ : θ → ω 2 θ, ω → ω 2 , order = 2 Note that τ σ 2 gives nothing new since τ σ 2 = στ . Similarly, σ 2 τ = τ σ. Thus σ 3 = τ 2 = 1, τ στ −1 = σ −1 (= σ 2 ).

(1)

At this point we have determined the multiplication table of G, but much more insight is gained by observing that (1) gives a presentation of S3 (Section 5.8, Problem 3). We conclude that G ∼ = S3 . The subgroups of G are {1}, G, < σ >, < τ >, < τ σ >, < τ σ 2 >

page 7 of Chapter 6

and the corresponding fixed fields are E, Q, Q(ω), Q(θ), Q(ωθ), Q(ω 2 θ). To show that the fixed field of < τ σ >= {1, τ σ} is Q(ωθ), note that < τ σ > has index 3 in G, so by the fundamental theorem, the corresponding fixed field has degree 3 over Q. Now τ σ takes ωθ to ω 2 ω 2 θ = ωθ and [Q(ωθ) : Q] = 3 (because the minimal polynomial of ωθ over Q is f). Thus Q(ωθ) is the entire fixed field. The other calculations are similar. Problems For Section 6.3 1. Suppose that E = F (α) is a finite Galois extension of F , where α is a root of the irreducible polynomial f ∈ F [X]. Assume that the roots of f are α1 = α, α2 , . . . , αn . Describe, as best you can from the given information, the Galois group of E/F . 2. Let E/Q be a finite Galois extension, and let x1 , . . . , xn be a basis for E over Q. Describe how you would find a primitive element, that is, an α ∈ E such that E = Q(α). (Your procedure need not be efficient.) 3. Let G be the Galois group of a separable irreducible polynomial f of degree n. Show that G is isomorphic to a transitive subgroup H of Sn . [Transitivity means that if i and j belong to {1, 2, . . . , n}, then for some σ ∈ H we have σ(i) = j. Equivalently, the natural action of H on {1, . . . , n}, given by h • x = h(x), is transitive.] 4. Use Problem 3 to determine the Galois group of an irreducible quadratic polynomial aX 2 + bX + c ∈ F [X], a 6= 0. Assume that the characteristic of F is not 2, so that the derivative of f is nonzero and f is separable. 5. Determine the Galois group of (X 2 − 2)(X 2 − 3) over Q. 6. In the Galois correspondence, suppose that Ki is the fixed field of the subgroup Hi , i = 1, 2. Identify the group corresponding to K = K1 ∩ K2. 7. Continuing Problem 6, identify the fixed field of H1 ∩ H2 . 8. Suppose that E is a splitting field of a separable polynomial f over F . Show that E/F is separable. [Since the extension is finite by (3.2.2) and normal by (3.5.7), E/F is Galois.] 9. Let G be the Galois group of f(X) = X 4 − 2 over Q. Thus if θ is the positive fourth root of 2, then G is the Galois group of Q(θ, i)/Q. Describe all 8 automorphisms in G. 10. Show that G is isomorphic to the dihedral group D8 . 11. Define σ(θ) = iθ, σ(i) = i, τ (θ) = θ, τ (i) = −i, as in the solution to Problem 10. Find the fixed field of the normal subgroup N = {1, στ, σ 2 , σ 3 τ } of G, and verify that the fixed field is a normal extension of Q. 6.4 Finite Fields Finite fields can be classified precisely. We will show that a finite field must have pn elements, where p is a prime and n is a positive integer. In addition, there is (up to isomorphism) only one finite field with pn elements. We sometimes use the notation GF (pn ) for this field; GF stands for “Galois field”. Also, the field with p elements will be denoted by Fp rather than Zp , to emphasize that we are working with fields. 6.4.1 Proposition Let E be a finite field of characteristic p. Then |E| = pn for some positive integer n. n Moreover, E is a splitting field for the separable polynomial f (X) = X p − X over Fp , so that any finite field with pn elements is isomorphic to E. Not only is E generated by the roots of f , but in fact E coincides with the set of roots of f . Proof. Since E contains a copy of Fp (see (2.1.3), Example 2), we may view E as a vector space over Fp . If the dimension of this vector space is n, then since each coefficient in a linear combination of basis vectors can be chosen in p ways, we have |E| = pn . n Now let E ∗ be the multiplicative group of nonzero elements of E. If α ∈ E ∗ , then αp −1 = 1 by pn Lagrange’s theorem, so α = α for every α ∈ E, including α = 0. Thus each element of E is a root of f , and f is separable by (3.4.5). Now f has at most pn distinct roots, and as we have already identified the pn elements of E as roots of f , in fact f has pn distinct roots and every root of f must belong to E. ♣ 6.4.2 Corollary If E is a finite field of characteristic p, then E/Fp is a Galois extension. The Galois group is cyclic and is generated by the Frobenius automorphism σ(x) = xp , x ∈ E.

page 8 of Chapter 6

Proof. E is a splitting field of a separable polynomial over Fp , so E/Fp is Galois; see (6.3.1). Since xp = x for each x ∈ Fp , Fp is contained in the fixed field F(< σ >). But each element of the fixed field is a root of X p − X, so F (< σ >) has at most p elements. Consequently, F (< σ >) = Fp . Now Fp = F (Gal(E/Fp ) by (6.1.2), so by the fundamental theorem, Gal(E/Fp ) =< σ >. ♣ 6.4.3 Corollary Let E/F be a finite extension of a finite field, with |E| = pn , |F | = pm . Then E/F is a Galois extension. Moreover, m divides n, and Gal(E/F ) is cyclic and is generated by the automorphism m τ (x) = xp , x ∈ E. Furthermore, F is the only subfield of E of size pm . Proof. If the degree of E/F is d, then as in (6.4.1), (pm )d = pn , so d = n/m and m|n. We may then m m reproduce the proof of (6.4.2) with Fp replaced by F , σ by τ, xp by xp , and X p by X p . Uniqueness of m F as a subfield of E with pm elements follows because there is only one splitting field over Fp for X p − X inside E; see (3.2.1). ♣. How do we know that finite fields (other than the Fp ) exist? There is no problem. Given any prime n p and positive integer n, we can construct E = GF (pn ) as a splitting field of X p − X over Fp . We have m just seen that if E contains a subfield F of size p , then m is a divisor of n. The converse is also true, as a consequence of the following basic result. 6.4.4 Theorem The multiplicative group of a finite field is cyclic. More generally, if G is a finite subgroup of the multiplicative group of an arbitrary field, then G is cyclic. Proof. G is a finite abelian group, hence contains an element g whose order r is the exponent of G, that is, the least common multiple of the orders of all elements of G; see Section 1.1, Problem 9. Thus if x ∈ G then the order of x divides r, so xr = 1. Therefore each element of G is a root of X r − 1, so |G| ≤ r. But |G| is a multiple of the order of every element, so |G| is at least as big as the least common multiple, so |G| ≥ r. We conclude that the order and the exponent are the same. But then g has order |G|, so G =< g > and G is cyclic. ♣ 6.4.5 Proposition GF (pm ) is a subfield of E = GF (pn ) if and only if m is a divisor of n. Proof. The “only if” part follows from (6.4.3), so assume that m divides n. If t is any positive integer greater than 1, then m|n iff (tm − 1)|(tn − 1). (A formal proof is not difficult, but I prefer to do an ordinary long division of tn − 1 by tm − 1. The successive quotients are tn−m , tn−2m , tn−3m , . . ., so the division will be successful iff n − rm = 0 for some positive integer r.) Taking t = p, we see that pm − 1 divides |E ∗ |, so by (6.4.4) and (1.1.4), E ∗ has a subgroup H of order pm − 1. By Lagrange’s theorem, each x ∈ H ∪ {0} m m satisfies xp = x. As in the proof of (6.4.1), H ∪ {0} coincides with the set of roots of X p − X. Thus we m may construct entirely inside GF (pn ) a splitting field of X p − X over Fp . But this splitting field is a copy of GF (pm ). ♣ In practice, finite fields are constructed by adjoining roots of carefully selected irreducible polynomials over Fp . The following result is very helpful. n

6.4.6 Theorem Let p be a prime and n a positive integer. Then X p − X is the product of all monic irreducible polynomials over Fp whose degree divides n. n

Proof. Let us do all calculations inside E = GF (pn ) = the set of roots of f (X) = X p − X. If g(X) is any monic irreducible factor of f (X), and deg g = m, then all roots of g lie in E. If α is any root of g, then Fp (α) is a finite field with pm elements, so m divides n by (6.4.5) or (6.4.3). Conversely, let g(X) be a monic irreducible polynomial over Fp whose degree m is a divisor of n. Then by (6.4.5), E contains a subfield with pm elements, and this subfield must be isomorphic to Fp (α). If β ∈ E corresponds to α under this isomorphism, then g(β) = 0 (because g(α) = 0) and f (β) = 0 (because β ∈ E). Since g is the minimal polynomial of β over Fp , it follows that g(X) divides f (X). By (6.4.1), the roots of f are distinct, so no irreducible factor can appear more than once. The theorem is proved. ♣ 6.4.7 The Explicit Construction of a Finite Field By (6.4.4), the multiplicative group E ∗ of a finite field E = GF (pn ) is cyclic, so E ∗ can be generated by a single element α. Thus E = Fp (α) = Fp [α], so that α is a primitive element of E. The minimal polynomial of α over Fp is called a primitive polynomial. The key point is that the nonzero elements of E are not

page 9 of Chapter 6

simply the nonzero polynomials of degree at most n − 1 in α, they are the powers of α. This is significant in applications to coding theory. Let’s do an example over F2 . The polynomial g(X) = X 4 + X + 1 is irreducible over F2 . One way to verify this is to factor X 16 − 1 = X + 1 over F2; the factors are the (necessarily monic) irreducible polynomials of degrees 1,2 and 4. To show that g is primitive, we compute powers of α: 16

α0 = 1, α1 = α, α2 = α2 , α3 = α3 , α4 = 1 + α (since g(α) = 0), α5 = α + α2 , α6 = α2 + α3 , α7 = α3 + α4 = 1 + α + α3 , α8 = α + α2 + α4 = 1 + α2 (since 1+1=0 in F2 ), α9 = α + α3 , α10 = 1 + α + α2 , α11 = α + α2 + α3 , α12 = 1 + α + α2 + α3 , α13 = 1 + α2 + α3 , α14 = 1 + α3 , and at this point we have all 24 − 1 = 15 nonzero elements of GF (16). The pattern now repeats, beginning with α15 = α + α4 = 1. For an example of a non-primitive polynomial, see Problem 1. Problems For Section 6.4 1. Verify that the irreducible polynomial X 4 + X 3 + X 2 + X + 1 ∈ F2 [X] is not primitive. 2. Let F be a finite field and d a positive integer. Show that there exists an irreducible polynomial of degree d in F [X]. 3. In (6.4.5) we showed that m|n iff (tm − 1)|(tn − 1) (t = 2, 3, . . .). Show that an equivalent condition is (X m − 1) divides (X n − 1). If E is a finite extension of a finite field, or more generally a finite separable extension of a field F , then by the theorem of the primitive element, E = F (α) for some α ∈ E. We now develop a condition equivalent to the existence of a primitive element. 4. Let E/F be a finite extension, with E = F (α) and F ≤ L ≤ E. Suppose that the minimal polynomial of Pr−1 α over L is g(X) = i=0 bi X i + X r , and let K = F (b0, . . . , br−1). If h is the minimal polynomial of α over K, show that g = h, and conclude that L = K. 5. Continuing Problem 4, show that there are only finitely many intermediate fields L between E and F . 6. Conversely, let E = F (α1 , . . . , αn ) be a finite extension with only finitely many intermediate fields between E and F . We are going to show by induction that E/F has a primitive element. If n = 1 there is nothing to prove, so assume the result holds for all integers less than n. If L = F (α1 , . . . , αn−1 ), show that E = F (β, αn ) for some β ∈ L. 7. Now assume (without loss of generality) that F is infinite. Show that there are distinct elements c, d ∈ F such that F (cβ + αn ) = F (dβ + αn ). 8. Continuing Problem 7, show that E = F (cβ + αn ). Thus a finite extension has a primitive element iff there are only finitely many intermediate fields. 9. Let α be an element of the finite field GF (pn ). Show that α and αp have the same minimal polynomial over Fp . 10. Suppose that α is an element of order 13 in the multiplicative group of nonzero elements in GF (3n ). Partition the integers {0, 1, . . . , 12} into disjoint subsets such that if i and j belong to the same subset, then αi and αj have the same minimal polynomial. Repeat for α an element of order 15 in GF (2n ). [Note that elements of the specified orders exist, because 13 divides 26 = 33 − 1 and 15 = 24 − 1.] 6.5 Cyclotomic Fields 6.5.1 Definitions and Comments Cyclotomic extensions of a field F are formed by adjoining nth roots of unity. Formally, a cyclotomic extension of F is a splitting field E of f (X) = X n − 1 over F . The roots of f are called nth roots of unity, and they form a multiplicative subgroup of the group E ∗ of nonzero elements of E. This subgroup must be cyclic by (6.4.4). A primitive nth root of unity is one whose order in E ∗ is n. It is tempting to say “obviously, primitive nth roots of unity must exist, just take a generator of the cyclic subgroup”. But suppose that F has characteristic p and p divides n, say n = mp. If ω is an nth root of unity, then 0 = ω n − 1 = (ω m − 1)p

page 10 of Chapter 6

so the order of ω must be less than n. To avoid this difficulty, we assume that the characteristic of F does not divide n. Then f 0 (X) = nX n−1 6= 0, so by (3.4.2), f is separable, and consequently E/F is Galois. Since there are n distinct nth roots of unity, there must be a primitive nth root of unity ω, and for any such ω, we have E = F (ω). If σ is any automorphism in the Galois group Gal(E/F ), then σ must take a primitive root of unity ω to another primitive root of unity ω r , where r and n are relatively prime. (See (1.1.5).) We can identify σ with r, and this shows that Gal(E/F ) is isomorphic to a subgroup of Un , the group of units mod n. Consequently, the Galois group is abelian. Finally, by the fundamental theorem (or (3.5.9)), [E : F ] = |Gal(E/F )|, which is a divisor of |Un | = ϕ(n). Cyclotomic fields are of greatest interest when the underlying field F is Q, the rational numbers, and from now on we specialize to that case. The primitive nth roots of unity are ei2πr/n where r and n are relatively prime. Thus there are ϕ(n) primitive nth roots of unity. Finding the minimal polynomial of a primitive nth root of unity requires some rather formidable equipment. 6.5.2 Definition The nth cyclotomic polynomial is defined by Y Ψn (X) = (X − ωi ) i

where the ωi are the primitive nth roots of unity in the field C of complex numbers. Thus the degree of Ψn (X) is ϕ(n). From the definition, we have Ψ1 (X) = X −1 and Ψ2 (X) = X +1. In general, the cyclotomic polynomials can be calculated by the following recursion formula, in which d runs through all positive divisors of n. 6.5.3 Proposition Xn − 1 =

Y

Ψd (X).

d|n

In particular, if p is prime, then Ψp (X) =

Xp − 1 = X p−1 + X p−2 + · · · + X + 1. X −1

Proof. If ω is an nthroot of unity, then its order in C∗ is a divisor d of n, and in this case, ω is a primitive dth root of unity, hence a root of Ψd (X). Conversely, if d|n, then any root of Ψd (X) is a dth, hence an nth , root of unity. ♣ From (6.5.3) we have Ψ3 (X ) = X 2 + X + 1, Ψ4 (X ) = X 2 + 1, Ψ5 (X ) = X 4 + X 3 + X 2 + X + 1, Ψ6 (X ) =

X 6 −1 (X−1)(X+1)(X 2 +X+1)

=

X 6 −1 (X 3 −1)(X+1)

X 3 +1 X+1

=

= X 2 − X + 1.

It is a natural conjecture that all coefficients of the cyclotomic polynomials are integers, and this turns out to be correct. 6.5.4 Proposition Ψn (X) ∈ Z[X]. Proof. By (6.5.3), we have Xn − 1 = [

Y

Ψd (X)]Ψn (X ).

d|n,d
By definition, the cyclotomic polynomials are monic, and by induction hypothesis, the expression in brackets is a monic polynomial in Z[X]. Thus Ψn (X) is the quotient of two monic polynomials with integer coefficients.

page 11 of Chapter 6

At this point, all we know for sure is that the coefficients of Ψn (X) are complex numbers. But if we apply ordinary long division, even in C, we know that the process will terminate, and this forces the quotient Ψn (X) to be in Z[X]. ♣ We now show that the nth cyclotomic polynomial is the minimal polynomial of each primitive nth root of unity. 6.5.5 Theorem Ψn (X) is irreducible over Q. Proof. Let ω be a primitive nth root of unity, with minimal polynomial f over Q. Since ω is a root of X n − 1, we have X n − 1 = f(X)g(X ) for some g ∈ Q[X]. Now it follows from (2.9.2) that if a monic polynomial over Z is the product of two monic polynomials f and g over Q, then in fact the coefficients of f and g are integers. If p is a prime that does not divide n, we will show that ω p is a root of f. If not, then it is a root of g. But g(ω p ) = 0 implies that ω is a root of g(X p ), so f(X) divides g(X p ), say g(X p ) = f(X)h(X). As above, h ∈ Z[X]. But by the binomial expansion modulo p, g(X)p ≡ g(X p ) = f (X )h(X ) mod p. Reducing the coefficients of a polynomial k(X ) mod p is equivalent to viewing it as an element k ∈ Fp [X], so we may write g(X)p = f (X)h(X). Then any irreducible factor of f must divide g, so f and g have a common factor. But then X n − 1 has a multiple root, contradicting (3.4.2). [This is where we use the fact that p does not divide n.] Now we claim that every primitive nthroot of unity is a root of f , so that deg f ≥ ϕ(n) =deg Ψn , and therefore f = Ψn by minimality of f . The best way to visualize this is via a concrete example with all the features of the general case. If ω is a primitive nthroot of unity where n = 175, then ω 72 is a primitive nth root of unity because 72 and 175 are relatively prime. Moreover, since 72 = 23 × 32 , we have ω 72 = (((((ω)2 )2 )2 )3)3 and the result follows. ♣ 6.5.6 Corollary The Galois group G of the nth cyclotomic extension Q(ω)/Q is isomorphic to the group Un of units mod n. Proof. By the fundamental theorem, |G| = [Q(ω) : Q] = deg Ψn = ϕ(n) = |Un |. Thus the isomorphism of G and a subgroup of Un (see (6.5.1)) is surjective. ♣ Problems For Section 6.5 1. If p is prime and p divides n, show that Ψpn (X ) = Ψn (X p ). (This formula is sometimes useful in computing the cyclotomic polynomials.) 2. Show that the group of automorphisms of a cyclic group of order n is isomorphic to the group Un of units mod n. (This can be done directly, but it is easier to make use of the results of this section.) We now do a detailed analysis of subgroups and intermediate fields associated with the cyclotomic extension Q7 = Q(ω)/Q where ω = ei2π/7 is a primitive 7th root of unity. The Galois group G consists of automorphisms σi , i = 1, 2, 3, 4, 5, 6, where σi (ω) = ω i . 3. Show that σ3 generates the cyclic group G. 4. Show that the subgroups of G are < 1 > (order 1), < σ6 > (order 2), < σ2 > (order 3), and G =< σ3 > (order 6). 5. The fixed field of < 1 > is Q7 and the fixed field of G is Q. Let K be the fixed field of < σ6 >. Show that ω + ω −1 ∈ K, and deduce that K = Q(ω + ω −1 ) = Q(cos 2π/7). 6. Let L be the fixed field of < σ2 >. Show that ω + ω 2 + ω 4 belongs to L but not to Q. 7. Show that L = Q(ω + ω 2 + ω 4 ). 8. If q = pr , p prime, show that Ψq (X) = tp−1 + tp−2 + · · · + 1 r−1

where t = X p . 9. Assuming that the first 6 cyclotomic polynomials are available [see after (6.5.3)], calculate Ψ18 (X) in an effortless manner.

page 12 of Chapter 6

6.6 The Galois Group of a Cubic Let f be a polynomial over F , with distinct roots x1 , . . . , xn in a splitting field E over F . The Galois group G of f permutes the xi , but which permutations belong to G? When f is a quadratic, the analysis is straightforward, and is considered in Section 6.3, Problem 4. In this section we look at cubics (and some other manageable cases), and the appendix to Chapter 6 deals with the quartic. 6.6.1 Definitions and Comments Let f be a polynomial with distinct roots x1 , . . . , xn as above. Define ∆(f ) =

Y

(xi − xj ).

i<j

The discriminant of f is defined by D(f ) = ∆2 =

Y (xi − xj )2 . i<j

√ Let’s look at a quadratic polynomial f (X) = X 2 + bX + c, with roots 12 (−b ± b2 − 4c). In order to divide by 2, we had better assume that the characteristic of F is not 2, and this assumption is usually made before defining the discriminant. In this case we have (x1 − x2 )2 = b2 − 4c, a familiar formula. Here are some basic properties of the discriminant. 6.6.2 Proposition Let E be a splitting field of the separable polynomial f over F , so that E/F is Galois. (a) D(f ) belongs to the base field F . (b) Let σ be an automorphism in the Galois group G of f . Then σ is an even permutation (of the roots of f ) iff σ(∆) = ∆, and σ is odd iff σ(∆) = −∆. (c) G ⊆ An , that is, G consists entirely of even permutations, iff D(f ) is the square of an element of F (for short, D ∈ F 2 ). Proof. Let us examine the effect of a transposition σ = (i, j) on ∆. Once again it is useful to consider a concrete example with all the features of the general case. Say n = 15, i = 7, j = 10. Then x3 − x7 → x3 − x10 , x3 − x10 → x3 − x7 x10 − x12 → x7 − x12 , x7 − x12 → x10 − x12 x7 − x8 → x10 − x8 , x8 − x10 → x8 − x7 x7 − x10 → x10 − x7 . The point of the computation is that the net effect of (i, j) on ∆ is to take xi − xj to its negative. Thus σ(∆) = −∆ when σ is a transposition. Thus if σ is any permutation, we have σ(∆) = ∆ if ∆ is even, and σ(∆) = −∆ if σ is odd. Consequently, σ(∆2 ) = (σ(∆))2 = ∆2 , so D belongs to the fixed field of G, which is F . This proves (a), and (b) follows because ∆ 6= −∆ (remember that the characteristic of F is not 2). Finally G ⊆ An iff σ(∆) = ∆ for every σ ∈ G iff ∆ ∈ F(G) = F . ♣ 6.6.3 The Galois Group of a Cubic In the appendix to Chapter 6, it is shown that the discriminant of the abbreviated cubic X 3 +pX +q is −4p3 −27q 2 , and the discriminant of the general cubic X 3 +aX 2 +bX +c is a2 (b2 − 4ac) − 4b3 − 27c2 + 18abc. Alternatively, the change of variable Y = X + criminant.

a 3

eliminates the quadratic term without changing the dis-

We now assume that the cubic polynomial f is irreducible as well as separable. Then the Galois group G is isomorphic to a transitive subgroup of S3 (see Section 6.3, Problem 3). By direct enumeration, G must be A3 or S3, and by (6.6.2(c)), G = A3 iff the discriminant D is a square in F .

page 13 of Chapter 6

If G = A3 , which is cyclic of order 3, there are no proper subgroups except {1}, so there are no intermediate fields strictly between E and F . However, if G = S3 , then the proper subgroups are {1, (2, 3)}, {1, (1, 3)}, {1, (1, 2)}, A3 = {1, (1, 2, 3), (1, 3, 2)}. If the roots of f are α1 , α2 and α3 , then the corresponding fixed fields are F (α1 ), F (α2 ), F (α3 ), F (∆) where A3 corresponds to F (∆) because only even permutations fix ∆. 6.6.4 Example Let f (X) = X 3 − 31X + 62 over Q. An application of the rational root test (Section 2.9, Problem 1) shows that f is irreducible. The discriminant is −4(−31)3 − 27(62)2 = 119164 − 103788 = 15376 = (124)2, which is a square in Q. Thus the Galois group of f is A3 . We now develop a result that can be applied to certain cubics, but which has wider applicability as well. The preliminary steps are also of interest. 6.6.5 Some Generating Sets of Sn (i) Sn is generated by the transpositions (1,2), (1,3),. . . , (1,n). [An arbitrary transposition (i, j) can be written as (1, i)(1, j)(1, i).] (ii) Sn is generated by transpositions of adjacent digits, i.e., (1, 2), (2, 3), . . . , (n − 1, n). [Since (1, j − 1)(j − 1, j)(1, j − 1) = (1, j), we have (1,2)(2,3)(1,2)=(1,3), (1,3)(3,4)(1,3)=(1,4), etc., and the result follows from (i).] (iii) Sn is generated by the two permutations σ1 = (1, 2) and τ = (1, 2, . . . , n). [If σ2 = τ σ1 τ −1 , then σ2 is obtained by applying τ to the symbols of σ1 (see Section 5.2, Problem 1). Thus σ2 = (2, 3). Similarly, σ3 = τ σ2 τ −1 = (3, 4), . . . , σn−1 = τ σn−2 τ −1 = (n − 1, n), and the result follows from (ii).] (iv) Sn is generated by (1,2) and (2,3,. . . ,n). [(1,2)(2,3,. . . ,n)=(1,2,3,. . . ,n), and (iii) applies.] 6.6.6 Lemma If f is an irreducible separable polynomial over F of degree n, and G is the Galois group of f , then n divides |G|. If n is a prime number p, then G contains a p-cycle. Proof. If α is any root of f , then [F (α) : F ] = n, so by the fundamental theorem, G contains a subgroup whose index is n. By Lagrange’s theorem, n divides |G|. If n = p, then by Cauchy’s theorem, G contains an element σ of order p. We can express σ as a product of disjoint cycles, and the length of each cycle must divide the order of σ. Since p is prime, σ must consist of disjoint p-cycles. But a single p-cycle already uses up all the symbols to be permuted, so σ is a p-cycle. ♣ 6.6.7 Proposition If f is irreducible over Q and of prime degree p, and f has exactly two nonreal roots in the complex field C, then the Galois group G of f is Sp . Proof. By (6.6.6), G contains a p-cycle σ. Now one of the elements of G must be complex conjugation τ , which is an automorphism of C that fixes R (hence Q). Thus τ permutes the two nonreal roots and leaves the p − 2 real roots fixed, so τ is a transposition. Since p is prime, σ k is a p-cycle for k = 1, . . . , p − 1. It follows that by renumbering symbols if necessary, we can assume that (1,2) and (1,2,. . . ,p) belong to G. By (6.6.5) part (iii), G = Sp . ♣ Problems For Section 6.6 In Problems 1-4, all polynomials are over the rational field Q, and in each case, you are asked to find the Galois group G.

page 14 of Chapter 6

1. f (X) = X 3 − 2 (do it two ways) 2. f (X) = X 3 − 3X + 1 3. f (X) = X 5 − 10X 4 + 2 4. f (X) = X 3 + 3X 2 − 2X + 1 (calculate the discriminant in two ways) 5. If f is a separable cubic, not necessarily irreducible, then there are other possibilities for the Galois group G of f besides S3 and A3 . What are they? 6. Let f be an irreducible cubic over Q with exactly one real root. Show that D(f ) < 0, and conclude that the Galois group of f is S3 . 7. Let f be an irreducible cubic√over Q with √ 3 distinct real roots. Show that D(f) > 0, so that the Galois group is A3 or S3 according as D ∈ Q or D ∈ /Q 6.7 Cyclic and Kummer Extensions The problem of solving a polynomial equation by radicals is thousands of years old, but it can be given a modern flavor. We are looking for roots of f ∈ F [X], and we are only allowed to use algorithms that do ordinary arithmetic plus the extraction of nthroots. The idea is to identify those polynomials whose roots √ can be found in this way. Now if a ∈ F and our algorithm computes θ = n a in some extension field of F , then θ is a root of X n − a, so it is natural to study splitting fields of X n − a. 6.7.1 Assumptions, Comments and a Definition Assume (i) E is a splitting field for f (X) = X n − a over F , where a 6= 0. (ii) F contains a primitive nth root of unity ω. These are natural assumption if we want to allow the computation of nth roots. If θ is any root of f in E, then the roots of f are θ, ωθ, . . . , ω n−1θ. (The roots must be distinct because a, hence θ, is nonzero.) Therefore E = F (θ). Since f is separable, the extension E/F is Galois (see (6.3.1)). If G =Gal(E/F ), then |G| = [E : F ] by the fundamental theorem (or by (3.5.9)). In general, a cyclic extension is a Galois extension whose Galois group is cyclic. 6.7.2 Theorem Under the assumptions of (6.7.1), E/F is a cyclic extension and the order of the Galois group G is a divisor of n. We have |G| = n if and only if f (X ) is irreducible over F . Proof. Let σ ∈ G; since σ permutes the roots of f by (3.5.1), we have σ(θ) = ω u(σ) θ. We identify integers u(σ) with the same residue mod n. If σi (θ) = ω u(σi ) θ, i = 1, 2, then σ1 (σ2 (θ)) = ω u(σ1 )+u(σ2 ) θ, so u(σ1 σ2 ) = u(σ1 ) + u(σ2) and u is a group homomorphism from G to Zn . If u(σ) is 0 mod n, then σ(θ) = θ, so σ is the identity and the homomorphism is injective. Thus G is isomorphic to a subgroup of Zn , so G is cyclic and |G| divides n. If f is irreducible over F , then |G| = [E : F ] = [F (θ) : F ] = deg f = n. If f is not irreducible over F , let g be a proper irreducible factor. If β is a root of g in E, then β is also a root of f , so E = F (β) and |G| = [E : F ] = [F (β) : F ] =deg g < n. ♣ Thus splitting fields of X n − a give rise to cyclic extensions. Conversely, we can prove that a cyclic extension comes from such a splitting field. 6.7.3 Theorem Let E/F be a cyclic extension of degree n, where F contains a primitive nthroot of unity ω. Then for some nonzero a ∈ F , f (X) = X n − a is irreducible over F and E is a splitting field for f over F. Proof. Let σ be a generator of the Galois group of the extension. By Dedekind’s lemma (6.1.6), the distinct automorphisms 1, σ, σ 2, . . . , σ n−1 are linearly independent over E. Thus 1 + ωσ + ω 2 σ 2 + · · · + ω n−1 σ n−1 is not identically 0, so for some β ∈ E we have θ = β + ωσ(β) + · · · + ω n−1 σ n−1 (β) 6= 0.

page 15 of Chapter 6

Now

σ(θ) = σ(β) + ωσ 2 (β) + · · · + ω n−2 σ n−1 (β) + ω n−1 σ n (β) = ω −1θ

since σ n (β) = β. We take a = θ n . To prove that a ∈ F , note that σ(θ n ) = (σ(θ))n = (ω −1 θ)n = θ n hence a belongs to the fixed field of Gal(E/F ), which is F . Now by definition of a, θ is a root of f(X) = X n − a, so the roots of X n − a are θ, ωθ, . . . , ω n−1 θ. Therefore F (θ) is a splitting field for f over F . Since σ(θ) = ω −1 θ, the automorphisms 1, σ, . . . , σ n−1 are still distinct when restricted to F (θ). Consequently, n ≤ |Gal(F (θ)/F )| = [F (θ) : F ] ≤ degf = n so [F (θ) : F ] = n. It follows that E = F (θ) and (since f must be the minimal polynomial of θ over F ) f is irreducible over F . ♣ A finite abelian group is a direct product of cyclic groups (or direct sum, in additive notation; see (4.6.4)). It is reasonable to expect that our analysis of cyclic Galois groups will help us to understand abelian Galois groups. 6.7.4 Definition A Kummer extension is a finite Galois extension with an abelian Galois group. 6.7.5 Theorem Let E/F be a finite extension, and assume that F contains a primitive nthroot of unity ω. Then E/F is a Kummer extension whose Galois group G has an exponent dividing n if and only if there are nonzero elements a1 ,√. . . , ar ∈ F such that E is a splitting field of (X n − a1 ) · · · (X n − ar ) over F . [For √ n short, E = F ( a1, . . . , n ar ).] Proof. We do the “if” part first. As in (6.7.1), we have E = F (θ1 , . . . , θr ) where θi is a root of X n − ai . If σ ∈ Gal(E/F ), then σ maps θi to another root of X n − ai , so σ(θi ) = ω ui (σ) θi . Thus if σ and τ are any two automorphisms in the Galois group G, then στ = τ σ and G is abelian. [The ui are integers, so ui (σ) + ui (τ ) = ui (τ ) + ui (σ).] Now restrict attention to the extension F (θi ). By (6.7.2), the Galois group of F (θi )/F has order dividing n, so σ n (θi ) = θi for all i = 1, . . . , r. Thus σ n is the identity, and the exponent of G is a divisor of n. For the “only if” part, observe that since G is a finite abelian group, it is a direct product of cyclic groups C1, . . . , Cr . For each i = 1, . . . , r, let Hi be the product of the Cj for j 6= i; by (1.5.3), Hi E G. We have G/Hi ∼ = Ci by the first isomorphism theorem. (Consider the projection mapping x1 · · · xr → xi ∈ Ci .) Let Ki be the fixed field of Hi . By the fundamental theorem, Ki /F is a Galois extension and its Galois group is isomorphic to G/Hi , hence isomorphic to Ci . Thus Ki /F is a cyclic extension of degree di = |Ci |, and di is a divisor of n. (Since G is the direct product of the Ci , some element of G has order di , so di divides the exponent of G and therefore divides n.) We want to apply (6.7.3) with n replaced by di , and this n/di is possible because F contains a primitive dth . We conclude that Ki = F (θi ), i root of unity, namely ω d (n/d ) n/d di i i i where θi is a nonzero element bi ∈ F . But θin = θi = bi = ai ∈ F . Finally, in the Galois correspondence, the intersection Trof the Hi is paired with the composite of the Ki , which is F (θ1 , . . . , θr ); see Section 6.3, Problem 7. But i=1 Hi = 1, so E = F (θ1 , . . . , θr ), and the result follows. ♣ Problems For Section 6.7

√ √ √ √ 1. Find the Galois group of the extension Q( 2, 3, 5, 7) [the splitting field of (X 2 − 2)(X 2 − 3)(X 2 − 5)(X 2 − 7)] over Q. 2. Suppose that E is a splitting field for f (X) = X n − a over F , a 6= 0, but we drop the second assumption in (6.7.1) that F contains a primitive nthroot of unity. Is it possible for the Galois group of E/F to be

page 16 of Chapter 6

cyclic? 3. Let E be a splitting field for X n − a over F , where a 6= 0, and assume that the characteristic of F does not divide n. Show that E contains a primitive nthroot of unity. We now assume that E is a splitting field for f (X) = X p − c over F , where c 6= 0, p is prime and the characteristic of F is not p. Let ω be a primitive pth root of unity in E (see Problem 3). Assume that f is not irreducible over F , and let g be an irreducible factor of f of degree d, where 1 ≤ d < p. Let θ be a root of g in E. 4. Let g0 be the product of the roots of g. (Since g0 is ± the constant term of g, g0 ∈ F .) Show that g0p = θ dp = cd . 5. Since d and p are relatively prime, there are integers a and b such that ad + bp = 1. Use this to show that if X p − c is not irreducible over F , then it must have a root in F . 6. Continuing Problem 5, show that if X p − c is not irreducible over F , then E = F (ω). 7. Continuing Problem 6, show that if X p − c is not irreducible over F , then X p − c splits over F if and only if F contains a primitive pth root of unity. Let E/F be a cyclic Galois extension of prime degree p, where p is the characteristic of F . Let σ be a generator of G =Gal(E/F ). It is a consequence of Hilbert’s Theorem 90 (see the Problems for Section 7.3) that there is an element θ ∈ E such that σ(θ) = θ + 1. Prove the Artin-Schreier theorem: 8. E = F (θ). 9. θ is a root of f (X) = X p − X − a for some a ∈ F . 10. f is irreducible over F (hence a 6= 0). Conversely, Let F be a field of prime characteristic p, and let E be a splitting field for f (X) = X p − X − a, where a is a nonzero element of F . 11. If θ is any root of f in E, show that E = F (θ) and that f is separable. 12. Show that every irreducible factor of f has the same degree d, where d = 1 or p. Thus if d = 1, then E = F , and if d = p, then f is irreducible over F . 13. If f is irreducible over F , show that the Galois group of f is cyclic of order p. 6.8 Solvability By Radicals 6.8.1 Definitions and Comments We wish to solve the polynomial equation f (X) = 0, f ∈ F [X], under the restriction that we are only allowed to perform ordinary arithmetic operations (addition, subtraction, multiplication and division) on the coefficients, along with extraction of nthroots (for any n = 2, 3, . . .). A sequence of operations of this type gives rise to a sequence of extensions F ≤ F (α1 ) ≤ F (α1 , α2 ) ≤ · · · ≤ F (α1 , . . . , αr ) = E where αn1 1 ∈ F and αni i ∈ F (α1, . . . , αi−1 ), i = 2, . . . , r. Equivalently, we have F = F0 ≤ F1 ≤ . . . ≤ Fr = E where Fi = Fi−1 (αi ) and αni i ∈ Fi−1 , i = 1, . . . , r. We say that E is a radical extension of F . It is convenient (and legal) to assume that n1 = · · · = nr = n. (Replace each ni by the product of all the ni . To justify this, observe that if αj belongs to a field L, then αmj ∈ L, m = 2, 3, . . ..) Unless otherwise specified, we will make this assumption in all hypotheses, conclusions and proofs. We have already seen three explicit classes of radical extensions: cyclotomic, cyclic and Kummer. (In the latter two cases, we assume that the base field contains a primitive nth root of unity.) We say that the polynomial f ∈ F [X] is solvable by radicals if the roots of f lie in some radical extension of F , in other words, there is a radical extension E of F such that f splits over E. Since radical extensions are formed by successively adjoining nth roots, it follows that the transitivity property holds: If E is a radical extension of F and L is a radical extension of E, then L is a radical extension of F .

page 17 of Chapter 6

A radical extension is always finite, but it need not be normal or separable. We will soon specialize to characteristic 0, which will force separability, and we can achieve normality by taking the normal closure (see (3.5.11)). 6.8.2 Proposition Let E/F be a radical extension, and let N be the normal closure of E over F . Then N/F is also a radical extension. Proof. E is obtained from F by successively adjoining α1 , . . . , αr , where αi is the nth root of an element in Fi−1 . On the other hand, N is obtained from F by adjoining not only the αi , but their conjugates αi1 , . . . , αim(i) . For any fixed i and j, there is an automorphism σ ∈ Gal(N/F ) such that σ(αi ) = αij (see (3.2.3), (3.5.5) and (3.5.6)). Thus n αij = σ(αi )n = σ(αni ) and since αni belongs to F (α1 , . . . , αi−1), it follows from (3.5.1) that σ(αni ) belongs to the splitting field Ki of Qi−1 n n j=1 min(αj , F ) over F . [Take K1 = F , and note that since α1 = b1 ∈ F , we have σ(α1 ) = σ(b1 ) = b1 ∈ F. n Alternatively, observe that by (3.5.1), σ must take a root of X − b1 to another root of this polynomial.] Thus we can display N as a radical extension of F by successively adjoining α11, . . . , α1m(1), . . . , αr1 , . . . , αrm(r) . ♣ 6.8.3 Preparation for the Main Theorem If F has characteristic 0, then a primitive nth root of unity ω can be adjoined to F to reach an extension F (ω); see (6.5.1). If E is a radical extension of F and F = F0 ≤ F1 ≤ · · · ≤ Fr = E, we can replace Fi by Fi (ω), i = 1, . . . , r, and E(ω) will be a radical extension of F . By (6.8.2), we can pass from E(ω) to its normal closure over F . Here is the statement we are driving at: Let f ∈ F [X], where F has characteristic 0. If f is solvable by radicals, then there is a Galois radical extension N = Fr ≥ · · · ≥ F1 ≥ F0 = F containing a splitting field K for f over F , such that each intermediate field Fi , i = 1, . . . , r, contains a primitive nth root of unity ω. We can assume that F1 = F (ω) and for i > 1, Fi is a splitting field for X n − bi over Fi−1 . [(Look at the end of the proof of (6.8.2).] By (6.5.1), F1 /F is a Kummer (Galois) extension, and by (6.7.2), each Fi /Fi−1 , i = 2, . . . , r is a cyclic (Galois) extension. We now do some further preparation. Suppose that K is a splitting field for f over F , and that the Galois group of K/F is solvable, with Gal(K/F ) = H0 D H1 D · · · D Hr = 1 with each Hi−1 /Hi abelian. By the fundamental theorem, we have the corresponding sequence of fixed fields F = K0 ≤ K1 ≤ · · · ≤ Kr = K with Ki /Ki−1 Galois and Gal(Ki /Ki−1 ) isomorphic to Hi−1 /Hi . Let us adjoin a primitive nthroot of unity ω to each Ki , so that we have fields Fi = Ki (ω) with F ≤ F0 ≤ F1 ≤ · · · ≤ Fr . We take n = |Gal(K/F )|. Since Fi can be obtained from Fi−1 by adjoining everything in Ki \ Ki−1 , we have Fi = Fi−1 Ki = Ki Fi−1 the composite of Fi−1 and Ki , i = 1, . . . , r. We may now apply Theorem 6.2.2. In the diamond diagram of Figure 6.2.1, at the top of the diamond we have Fi , on the left Ki , on the right Fi−1 , and on the bottom Ki ∩ Fi−1 ⊇ Ki−1 (see Figure 6.8.1). We conclude that Fi /Fi−1 is Galois, with a Galois group isomorphic to a subgroup of Gal(Ki /Ki−1 ). Since Gal(Ki /Ki−1 ) ∼ = Hi−1 /Hi , it follows that Gal(Fi /Fi−1)

page 18 of Chapter 6

is abelian. Moreover, the exponent of this Galois group divides the order of H0 , which coincides with the size of Gal(K/F ). (This explains our choice of n.)

6.8.4 Galois’ Solvability Theorem Let K be a splitting field for f over F , where F has characteristic 0. Then f is solvable by radicals if and only if the Galois group of K/F is solvable. Proof. If f is solvable by radicals, then as in (6.8.3), we have F = F0 ≤ F1 ≤ · · · ≤ Fr = N where N/F is Galois, N contains a splitting field K for f over F , and each Fi /Fi−1 is Galois with an abelian Galois group. By the fundamental theorem, the corresponding sequence of subgroups is 1 = Hr E Hr−1 E · · · E H0 = G = Gal(N/F ) with each Hi−1 /Hi abelian. Thus G is solvable, and since Gal(N/F ) Gal(K/F ) ∼ = Gal(N/K) [map Gal(N/F ) → Gal(K/F ) by restriction; the kernel is Gal(N/K)], Gal(K/F ) is solvable by (5.7.4). Conversely, assume that Gal(K/F ) is solvable. Again as in (6.8.3), we have F ≤ F0 ≤ F1 ≤ · · · ≤ Fr where K ≤ Fr , each Fi contains a primitive nth root of unity, with n = |Gal(K/F )|, and Gal(Fi /Fi−1 ) is abelian with exponent dividing n for all i = 1, . . . , r. Thus each Fi /Fi−1 is a Kummer extension whose Galois group has an exponent dividing n. By (6.7.5) (or (6.5.1) for the case i = 1), each Fi /Fi−1 is a radical extension. By transitivity (see (6.8.1)), Fr is a radical extension of F . Since K ⊆ Fr , f is solvable by radicals. ♣ 6.8.5 Example Let f (X ) = X 5 − 10X 4 + 2 over the rationals. The Galois group of f is S5 , which is not solvable. (See Section 6.6, Problem 3 and Section 5.7, Problem 5.) Thus f is not solvable by radicals. There is a fundamental idea that needs to be emphasized. The significance of Galois’ solvability theorem is not simply that there are some examples of bad polynomials. The key point is there is no general method for solving a polynomial equation over the rationals by radicals, if the degree of the polynomial is 5 or more. If there were such a method, then in particular it would work on Example (6.8.5), a contradiction. Problems For Section 6.8 In the exercises, we will sketch another classical problem, that of constructions with ruler and compass. In Euclidean geometry, we start with two points (0,0) and (1,0), and we are allowed the following constructions.

page 19 of Chapter 6

(i) Given two points P and Q, we can draw a line joining them; (ii) Given a point P and a line L, we can draw a line through P parallel to L; (iii) Given a point P and a line L, we can draw a line through P perpendicular to L; (iv) Given two points P and Q, we can draw a circle with center at P passing through Q; (v) Let A, and similarly B, be a line or a circle. We can generate new points, called constructible points, by forming the intersection of A and B. If (c, 0) (equivalently (0, c)) is a constructible point, we call c a constructible number. It follows from (ii) and (iii) that (a, b) is a constructible point iff a and b are constructible numbers. It can be shown that every rational number is constructible, and that the constructible numbers form a field. Now in (v), the intersection of A and B can be found by ordinary arithmetic plus at worst the extraction of a square root. Therefore c is constructible iff there are fields Q = F0 ≤ F1 · · · ≤ Fr such that c ∈ Fr and each [Fi : Fi−1] is 1 or 2. Thus if c is constructible, then c is algebraic over Q and [Q(c) : Q] is a power of 2. 1. (Trisecting the angle) If it is possible to trisect any angle with ruler and compass, then in particular a 60 degree angle can be trisected, so that α = cos 20◦ is constructible. Using the identity ei3θ = cos 3θ + i sin 3θ = (cos θ + i sin θ)3 , reach a contradiction. 2. (Duplicating the cube) Show that it is impossible to√construct, with ruler and compass, a cube whose volume is exactly 2. (The side of such a cube would be 3 2.) 3. (Squaring the circle) Show that if it were possible to construct a square with area π, then π would be algebraic over Q. (It is known that π is transcendental over Q.) To construct a regular n-gon, that is, a regular polygon with n sides, n ≥ 3,we must be able to construct an angle of 2π/n; equivalently, cos 2π/n must be a constructible number. Let ω = ei2π/n , a primitive nth root of unity. 4. Show that [Q(ω) : Q(cos 2π/n)] = 2. 5. Show that if a regular n-gon is constructible, then the Euler phi function ϕ(n) is a power of 2. Conversely, assume that ϕ(n) is a power of 2. 6. Show that Gal(Q(cos 2π/n)/Q) is a 2-group, that is, a p-group with p = 2. 7. By Section 5.7, Problem 7, every nontrivial finite p-group has a subnormal series in which every factor has order p. Use this (with p = 2) to show that a regular n-gon is constructible. 8. From the preceding, a regular n-gon is constructible if and only if ϕ(n) is a power of 2. Show that an equivalent condition is that n = 2s q1 · · · qt , s, t = 0, 1, . . ., where the qi are distinct Fermat primes, that is, primes of the form 2m + 1 for some positive integer m. 9. Show that if 2m + 1 is prime, then m must be a power of 2. The only known Fermat primes have m = 2a , where a = 0, 1, 2, 3, 4 (232 + 1 is divisible by 641). [The key point is that if a is odd, then X + 1 divides X a + 1 in Z[X ]; the quotient is X a−1 − X a−2 + · · · − X + 1 (since a − 1 is even).] Let F be the field of rational functions in n variables e1 , . . . , en over a field k with characteristic 0, and let f (X) = X n − e1 X n−1 + e2 X n−2 − · · · + (−1)n e0 ∈ F [X]. If α1 , . . . , αn are the roots of f in a splitting field over F , then the ei are the elementary symmetric functions of the αi . Let E = F (α1, . . . , αn ), so that E/F is a Galois extension and G = Gal(E/F ) is the Galois group of f. 10. Show that G ∼ = Sn . 11. What can you conclude from Problem 10 about solvability of equations? 6.9 Transcendental Extensions 6.9.1 Definitions and Comments An extension E/F such that at least one α ∈ E is not algebraic over F is said to be transcendental. An idea analogous to that of a basis of an arbitrary vector space V turns out to be profitable in studying transcendental extensions. A basis for V is a subset of V that is linearly independent and spans V . A key result, whose proof involves the Steinitz exchange, is that if {x1 , . . . , xm } spans V and S is a linearly independent subset of V , then |S| ≤ m. We are going to replace linear independence by algebraic independence and spanning by algebraic spanning. We will find that every transcendental extension has a

page 20 of Chapter 6

transcendence basis, and that any two transcendence bases for a given extension have the same cardinality. All these terms will be defined shortly. The presentation in the text will be quite informal; I believe that this style best highlights the strong connection between linear and algebraic independence. An indication of how to formalize the development is given in a sequence of exercises. See also Morandi, “Fields and Galois Theory”, pp.173-182. Let E/F be an extension. The elements t1 , . . . , tn ∈ E are algebraically dependent over F (or the set {t1 , . . . , tn } is algebraically dependent over F ) if there is a nonzero polynomial f ∈ F [X1 , . . . , Xn ] such that f (t1 , . . . , tn ) = 0; otherwise the ti are algebraically independent over F . Algebraic independence of an infinite set means algebraic independence of every finite subset. Now if a set T spans a vector space V , then each x in V is a linear combination of elements of T , so that x depends on T in a linear fashion. Replacing “linear” by “algebraic”, we say that the element t ∈ E depends algebraically on T over F if t is algebraic over F (T ), the field generated by T over F (see Section 3.1, Problem 1). We say that T spans E algebraically over F if each t in E depends algebraically on T over F , that is, E is an algebraic extension of F (T ). A transcendence basis for E/F is a subset of E that is algebraically independent over F and spans E algebraically over F . (From now on, we will frequently regard F as fixed and drop the phrase “over F ”.) 6.9.2 Lemma If S is a subset of E, the following conditions are equivalent. (i) S is a transcendence basis for E/F ; (ii) S is a maximal algebraically independent set; (iii) S is a minimal algebraically spanning set. Thus by (ii), S is a transcendence basis for E/F iff S is algebraically independent and E is algebraic over F (S). Proof. (i) implies (ii) If S ⊂ T where T is algebraically independent, let u ∈ T \ S. Then u cannot depend on S algebraically (by algebraic independence of T ), so S cannot span E algebraically. (ii) implies (i) If S does not span E algebraically, then there exists u ∈ E such that u does not depend algebraically on S. But then S ∪ {u} is algebraically independent, contradicting maximality of S. (i) implies (iii) If T ⊂ S and T spans E algebraically, let u ∈ S \ T . Then u depends algebraically on T , so T ∪ {u}, hence S, is algebraically dependent , a contradiction. (iii) implies (i) If S is algebraically dependent, then some u ∈ S depends algebraically on T = S \ {u}. But then T spans E algebraically, a contradiction. ♣ 6.9.3 Proposition Every transcendental extension has a transcendence basis. Proof. The standard argument via Zorn’s lemma that an arbitrary vector space has a maximal linearly independent set (hence a basis) shows that an arbitrary transcendental extension has a maximal algebraically independent set, which is a transcendence basis by (6.9.2). ♣ For completeness, if E/F is an algebraic extension, we can regard ∅ as a transcendence basis. 6.9.4 The Steinitz Exchange If {x1 , . . . , xm } spans E algebraically and S is algebraically independent, then |S| ≤ m. Proof. Suppose that S has at least m + 1 elements y1 , . . . , ym+1 . Since the xi span E algebraically, y1 depends algebraically on x1 , . . . , xm . The algebraic dependence relation must involve at least one xi , say x1 . (Otherwise, S would be algebraically dependent.) Then x1 depends algebraically on y1 , x2 , . . . , xm , so {y1 , x2 , . . . , xm } spans E algebraically. We claim that for every i = 1, . . . , m, {y1 , . . . , yi , xi+1 , . . . , xm } spans E algebraically. We have just proved the case i = 1. If the result holds for i, then yi+1 depends algebraically on {y1 , . . . , yi , xi+1 , . . . , xm }, and the dependence relation must involve at least one xj , say xi+1 for convenience. (Otherwise, S would be algebraically dependent.) Then xi+1 depends algebraically on y1, . . . , yi+1 , xi+2 , . . . , xm , so {y1, . . . , yi+1 , xi+2 , . . . , xm } spans E algebraically, completing the induction. Since there are more y’s than x’s, eventually the x’s disappear, and y1 , . . . , ym span E algebraically. But then ym+1 depends algebraically on y1 , . . . , ym , contradicting the algebraic independence of S. ♣

page 21 of Chapter 6

6.9.5 Corollary Let S and T be transcendence bases of E. Then either S and T are both finite or they are both infinite; in the former case, |S| = |T |. Proof. Assume that one of the transcendence bases, say T , is finite. By (6.9.4), |S| ≤ |T |, so S is finite also. By a symmetrical argument, |T | ≤ |S|, so |S| = |T |. ♣ 6.9.6 Proposition If S and T are arbitrary transcendence bases for E, then |S| = |T |. [The common value is called the transcendence degree of E/F .] Proof. By (6.9.5), we may assume that S and T are both infinite. Let T = {yi : i ∈ I}. If x ∈ S, then x depends algebraically on finitely many elements yi1 , . . . , yir in T . Define I(x) to be the set of indices {i1 , . . . , ir }. It follows that I = ∪{I(x) : x ∈ S}. For if j belongs to none of the I(x), then we can remove yj from T and the resulting set will still span E algebraically, contradicting (6.9.2) part (iii). Now an element of ∪{I(x) : x ∈ S} is determined by selecting an element x ∈ S and then choosing an index in I(x). Since I(x) is finite, we have |I(x)| ≤ ℵ0 . Thus |I| = |

[ {I(x) : x ∈ S}| ≤ |S|ℵ0 = |S|

since S is infinite. Thus |T | ≤ |S|. By symmetry, |S| = |T |. ♣ 6.9.7 Example Let E = F (X1 , . . . , Xn ) be the field of rational functions in the variables X1 , . . . , Xn with coefficients in F . If f (X1 , . . . , Xn ) = 0, then f is the zero polynomial, so S = {X1 , . . . , Xn } is an algebraically independent set. Since E = F (S), E is algebraic over F (S) and therefore S spans E algebraically. Thus S is a transcendence basis. Now let T = {X1u1 , . . . , Xnun }, where u1 , . . . , un are arbitrary positive integers. We claim that T is also a transcendence basis. As above, T is algebraically independent. Moreover, each Xi is algebraic over F (T ). To see what is going on, look at a concrete example, say T = {X15 , X23 , X34 }. If f (Z) = Z 3 − X23 ∈ F (T )[Z], then X2 is a root of f , so X2 , and similarly each Xi , is algebraic over F (T ). By (3.3.3), E is algebraic over F (T ), so T is a transcendence basis. Problems For Section 6.9 1. If S is an algebraically independent subset of E over F , T spans E algebraically over F , and S ⊆ T , show that there is a transcendence basis B such that S ⊆ B ⊆ T . 2. Show that every algebraically independent set can be extended to a transcendence basis, and that every algebraically spanning set contains a transcendence basis. 3. Prove carefully, for an extension E/F and a subset T = {t1, . . . , tn } ⊆ E, that the following conditions are equivalent. (i) T is algebraically independent over F ; (ii) For every i = 1, . . . , n, ti is transcendental over F (T \ {ti }); (iii) For every i = 1, . . . , n, ti is transcendental over F (t1 , . . . , ti−1 ) (where the statement for i = 1 is that t1 is transcendental over F ). 4. Let S be a subset of E that is algebraically independent over F . Show that if t ∈ E, then t is transcendental over F (S) if and only if S ∪ {t} is algebraically independent over F . [Problems 3 and 4 suggest the reasoning that is involved in formalizing the results of this section.] 5. Let F ≤ K ≤ E, with S a subset of K that is algebraically independent over F , and T a subset of E that is algebraically independent over K. Show that S ∪ T is algebraically independent over F , and S ∩ T = ∅. 6. Let F ≤ K ≤ E, with S a transcendence basis for K/F and T a transcendence basis for E/K. Show that S ∪ T is a transcendence basis for E/F . Thus if tr deg abbreviates transcendence degree, then by Problem 5, tr deg(E/F ) = tr deg(K/F ) + tr deg(E/K). 7. Let E be an extension of F , and T = {t1 , . . . , tn } a finite subset of E. Show that F (T ) is F -isomorphic to the rational function field F (X1 , . . . , Xn ) if and only if T is algebraically independent over F . 8. An algebraic function field F in one variable over K is a field F/K such that there exists x ∈ F

page 22 of Chapter 6

transcendental over K with [F : K(x)] < ∞. If z ∈ F , show that z is transcendental over K iff [F : K(z)] < ∞. 9. Find the transcendence degree of the complex field over the rationals. Appendix To Chapter 6 We will develop a method for calculating the discriminant of a polynomial and apply the result to a cubic. We then calculate the Galois group of an arbitrary quartic. A6.1 Definition If x1 , . . . , xn (n ≥ 2) are arbitrary elements of a field, the Vandermonde determinant of the xi is ¯ ¯ 1 ··· 1 ¯ ¯ 1 ¯ ¯ x2 ··· xn ¯ ¯ x1 ¯ ¯ det V = ¯ .. ¯ . ¯ ¯ ¯ n−1 ¯ n−1 n−1 x1 x2 · · · xn A6.2 Proposition det V =

Y

(xj − xi ).

i<j

¡n¢ Proof. Q det V is a polynomial h of degree 1 + 2 + · · · + (n − 1) = 2 in the variables x1 , . . . , xn , as is g = i<j (xj − xi ). If xi = xj for i < j, then the determinant is 0, so by the remainder theorem (2.5.2), each factor of g, hence g itself, divides h. Since h and g have the same degree, h = cg for some constant c. Now look at the leading terms of h and g, i.e., those terms in which xn appears to as high a power as possible, and subject to this constraint, xn−1 appears to as high a power as possible, etc. In both cases, the leading term is x2 x23 · · · xn−1 , and therefore c must be 1. (For this step it is profitable to regard the xi as n abstract variables in a polynomial ring. Then monomials xr11 · · · xrnn with different sequences (r1 , . . . , rn ) of exponents are linearly independent.) ♣ A6.3 Corollary If f is a polynomial in F [X] with roots x1 , . . . , xn in some splitting field over F , then the discriminant of f is (det V )2 . Proof. By definition of the discriminant D of f (see 6.6.1), we have D = ∆2 where ∆ = ± det V . ♣ A6.4 Computation of the Discriminant det(V V t ), which is the 1 x ··· 1 1 ··· xn   1 x2  ..  . · · · xn−1 1 x n n

The square of the determinant of V is  1 1 x2  x1  ..  . n−1 xn−1 x 1 2 and this in turn is

¯ ¯ t0 ¯ ¯ t1 ¯ ¯ ¯ ¯ tn−1

t1 t2 .. .

··· ···

tn

···

determinant of · · · xn−1  1

···

xn−1  2  

...

xn−1 n

¯ tn−1 ¯ ¯ tn ¯ ¯ ¯ ¯ ¯ t2n−2

where the power sums tr are given by t0 = n, tr =

n X

xri , r ≥ 1.

i=1

We must express the power sums in terms of the coefficients of the polynomial f . This will involve, improbably, an exercise in differential calculus. We have F (z) =

n Y

(1 − xi z) =

i=1

n X i=0

ci z i with c0 = 1;

page 23 of Chapter 6

the variable z ranges over real numbers. Take the logarithmic derivative of F to obtain ∞

∞

X −xi X X j+1 X F 0 (z) d = log F (z) = =− xi z j = − tj+1 z j . F (z) dz 1 − xi z n

n

i=1

i=1 j=0

Thus F 0 (z) + F (z)

∞ X

j=0

tj+1 z j = 0,

j=0

that is,

n X

ici z i−1 +

i=1

n X

ci z i

i=0

∞ X

tj z j−1 = 0.

j=1

Equating powers of z r−1 , we have, assuming that n ≥ r, rcr + c0 tr + c1tr−1 + · · · + cr−1 t1 = 0;

(1)

if r > n, the first summation does not contribute, and we get tr + c1 tr−1 + · · · + cn tr−n = 0.

(2)

Our situation Pn is a bit awkward here because the roots of F (z) are the reciprocals of the xi . The xi are the roots of i=0 ai z i where ai = cn−i (so that an = c0 = 1). The results can be expressed as follows. Pn A6.5 Newton’s Identities If f (X) = i=0 ai X i (with an = 1) is a polynomial with roots x1 , . . . , xn , then the power sums ti satisfy tr + an−1 tr−1 + · · · + an−r+1 t1 + ran−r = 0, r ≤ n

(3)

tr + an−1 tr−1 + · · · + a0 tr−n = 0, r > n.

(4)

and A6.6 The Discriminant of a Cubic First consider the case where the X 2 term is missing, so that f (X) = X 3 + pX + q. Then n = t0 = 3, a0 = q, a1 = p, a2 = 0 (a3 = 1). Newton’s identities yield t1 + a2 = 0, t1 = 0; t2 + a2 t1 + 2a1 = 0, t2 = −2p; t3 + a2 t2 + a1 t1 + 3a0 = 0, t3 = −3a0 = −3q; t4 + a2 t3 + a1 t2 + a0 t1 = 0, t4 = −p(−2p) = 2p2 ¯ ¯ 3 0 ¯ D = ¯¯ 0 −2p ¯ −2p −3q

¯ −2p ¯¯ −3q ¯¯ = −4p3 − 27q 2 . 2p2 ¯

We now go to the general case f (X) = X 3 + aX 2 + bX + c. The quadratic term can be eliminated by the substitution Y = X + a3 . Then a a a f(X) = g(Y ) = (Y − )3 + a(Y − )2 + b(Y − ) + c 3 3 3 a2 2a3 ba ,q = − + c. 3 27 3 Since the roots of f are translations of the roots of g by the same constant, the two polynomials have the same discriminant. Thus D = −4p3 − 27q 2 , which simplifies to = Y 3 + pY + q where p = b −

D = a2(b2 − 4ac) − 4b3 − 27c2 + 18abc.

page 24 of Chapter 6

We now consider the Galois group of a quartic X 4 + aX 3 + bX 2 + cX + d, assumed irreducible and separable over a field F . As above, the translation Y = X + a4 eliminates the cubic term without changing the Galois group, so we may assume that f (X) = X 4 + qX 2 + rX + s. Let the roots of f be x1 , x2 , x3 , x4 (distinct by separability), and let V be the four group, realized as the subgroup of S4 containing the permutations (1,2)(3,4), (1,3)(2,4) and (1,4)(2,3), along with the identity. By direct verification (i.e., brute force), V E S4 . If G is the Galois group of f (regarded as a group of permutations of the roots), then V ∩ G E G by the second isomorphism theorem. A6.7 Lemma F (V ∩ G) = F (u, v, w), where u = (x1 + x2 )(x3 + x4 ), v = (x1 + x3 )(x2 + x4 ), w = (x1 + x4 )(x2 + x3 ). Proof. Any permutation in V fixes u, v and w, so GF (u, v, w) ⊇ V ∩ G. If σ ∈ G but σ ∈ / V ∩ G then (again by direct verification) σ moves at least one of u, v, w. For example, (1,2,3) sends u to w, and (1,2) sends v to w. Thus σ ∈ / GF (u, v, w). Therefore GF (u, v, w) = V ∩ G, and an application of the fixed field operator F completes the proof. ♣ A6.8 Definition The resolvent cubic of f (X) = X 4 + qX 2 + rX + s is g(X) = (X − u)(X − v)(X − w). To compute g, we must express its coefficients in terms of q, r and s. First note that u − v = −(x1 − x4 )(x2 − x3), u − w = −(x1 − x3 )(x2 − x4 ), v − w = −(x1 − x2 )(x3 − x4 ). Thus f and g have the same discriminant. Now X 4 + qX 2 + rX + s = (X 2 + kX + l)(X 2 − kX + m) where the appearance of k and −k is explained by the missing cubic term. Equating coefficients gives l + m − k 2 = q, k(m − l) = r, lm = s. Solving the first two equations for m and adding, we have 2m = k 2 + q + r/k, and solving the first two equations for l and adding, we get 2l = k 2 + q − r/k. Multiply the last two equations and use lm = s to get a cubic in k 2 , namely k 6 + 2qk 4 + (q 2 − 4s)k 2 − r2 = 0. (This gives a method for actually finding the roots of a quartic.) To summarize, f(X) = (X 2 + kX + l)(X 2 − kX + m) where k 2 is a root of h(X ) = X 3 + 2qX 2 + (q 2 − 4s)X − r2 . We claim that the roots of h are simply −u, −v, −w. For if we arrange the roots of f so that x1 and x2 are the roots of X 2 + kX + l, and x3 and x4 are the roots of X 2 − kX + m, then k = −(x1 + x2 ), −k = −(x3 + x4), so −u = k 2 . The argument for −v and −w is similar. Therefore to get g from h, we simply change the sign of the quadratic and constant terms, and leave the linear term alone. A6.9 An Explicit Formula For The Resolvent Cubic: g(X) = X 3 − 2qX 2 + (q 2 − 4s)X + r2 . We need some results concerning subgroups of Sn , n ≥ 3. A6.10 Lemma (i) An is generated by 3-cycles, and every 3-cycle is a commutator. (ii) The only subgroup of Sn with index 2 is An . Proof. For the first assertion of (i), see Section 5.6, Problem 4. For the second assertion of (i), note that (a, b)(a, c)(a, b)−1(a, c)−1 = (a, b)(a, c)(a, b)(a, c) = (a, b, c).

page 25 of Chapter 6

To prove (ii), let H be a subgroup of Sn with index 2; H is normal by Section 1.3, Problem 6. Thus Sn /H has order 2, hence is abelian. But then by (5.7.2), part 5, Sn0 ≤ H, and since An also has index 2, the same argument gives Sn0 ≤ An . By (i), An ≤ Sn0 , so An = Sn0 ≤ H. Since An and H have the same finite number of elements n!/2, it follows that H = An . ♣ A6.11 Proposition Let G be a subgroup of S4 whose order is a multiple of 4, and let V be the four group (see the discussion preceding A6.7). Let m be the order of the quotient group G/(G ∩ V ). Then (a) If m = 6, then G = S4 ; (b) If m = 3, then G = A4 ; (c) If m = 1, then G = V ; (d) If m = 2, then G = D8 or Z4 or V ; (e) If G acts transitively on {1, 2, 3, 4}, then the case G = V is excluded in (d). [In all cases, equality is up to isomorphism.] Proof. If m = 6 or 3, then since |G| = m|G ∩ V |, 3 is a divisor of |G|. By hypothesis, 4 is also a divisor, so |G| is a multiple of 12. By A6.10 part (ii), G must be S4 or A4. But |S4 /(S4 ∩ V )| = |S4 /V | = 24/4 = 6 and |A4 /(A4 ∩ V )| = |A4 /V | = 12/4 = 3 proving both (a) and (b). If m = 1, then G = G ∩ V , so G ≤ V , and since |G| is a multiple of 4 and |V | = 4, we have G = V , proving (c). If m = 2, then |G| = 2|G ∩ V |, and since |V | = 4, |G ∩ V | is 1,2 or 4. If it is 1, then |G| = 2 × 1 = 2, contradicting the hypothesis. If it is 2, then |G| = 2 × 2 = 4, and G = Z4 or V (the only groups of order 4). Finally, assume |G ∩ V | = 4, so |G| = 8. But a subgroup of S4 of order 8 is a Sylow 2-subgroup, and all such subgroups are conjugate and therefore isomorphic. One of these subgroups is D8 , since the dihedral group of order 8 is a group of permutations of the 4 vertices of a square. This proves (d). If m = 2, G acts transitively on {1, 2, 3, 4} and |G| = 4, then by the orbit-stabilizer theorem, each stabilizer subgroup G(x) is trivial (since there is only one orbit, and its size is 4). Thus every permutation in G except the identity moves every integer 1,2,3,4. Since |G ∩ V | = 2, G consists of the identity, one other element of V , and two elements not in V , which must be 4-cycles. But a 4-cycle has order 4, so G must be cyclic, proving (e). ♣ A6.12 Theorem Let f be an irreducible separable quartic, with Galois group G. Let m be the order of the Galois group of the resolvent cubic. Then: (a) If m = 6, then G = S4 ; (b) If m = 3, then G = A4 ; (c) If m = 1, then G = V ; (d) If m = 2 and f is irreducible over L = F (u, v, w), where u, v and w are the roots of the resolvent cubic, then G = D8 ; (e) If m = 2 and f is reducible over L, then G = Z4 . Proof. By A6.7 and the fundamental theorem, [G : G ∩ V ] = [L : F ]. Now the roots of the resolvent cubic g are distinct, since f and g have the same discriminant. Thus L is a splitting field of a separable polynomial, so L/F is Galois. Consequently, [L : F ] = m by (3.5.9). To apply (A6.11), we must verify that |G| is a multiple of 4. But this follows from the orbit-stabilizer theorem: since G acts transitively on the roots of f , there is only one orbit, of size 4 = |G|/|G(x)|. Now (A6.11) yields (a), (b) and (c), and if m = 2, then G = D8 or Z4 . To complete the proof, assume that m = 2 and G = D8 . Thinking of D8 as the group of symmetries of a square with vertices 1,2,3,4, we can take D8 to be generated by (1,2,3,4) and (2,4), with V = {1, (1, 2)(3, 4), (1, 3)(2, 4), (1, 4)(2, 3)}. The elements of V are symmetries of the square, hence belong to D8 ; thus V = G ∩ V = Gal(E/L) by (A6.7). [E is a splitting field for f over F .] Since V is transitive, for each i, j = 1, 2, 3, 4, i 6= j, there is an L-automorphism τ of E such that τ (xi ) = xj . Applying τ to the

page 26 of Chapter 6

equation h(xi ) = 0, where h is the minimal polynomial of xi over L, we see that each xj is a root of h, and therefore f |h. But h|f by minimality of h, so h = f , proving that f is irreducible over L. Finally, assume m = 2 and G = Z4, which we take as {1, (1, 2, 3, 4), (1, 3)(2, 4), (1, 4, 3, 2)}. Then G ∩ V = {1, (1, 3)(2, 4)}, which is not transitive. Thus for some i 6= j, xi and xj are not roots of the same irreducible polynomial over L. In particular, f is reducible over L. ♣ A6.13 Example Let f (X) = X 4 + 3X 2 + 2X + 1 over Q, with q = 3, r = 2, s = 1. The resolvent cubic is, by (A6.9), g(X) = X 3 − 6X 2 + 5X + 4. To calculate the discriminant of g, we can use the general formula in (A6.6), or compute g(X + 2) = (X + 2)3 − 6(X + 2)2 + 5(X + 2) + 4 = X 3 − 7X − 2. [The rational root test gives irreducibility of g and restricts a factorization of f to (X 2 + aX ± 1)(X 2 − aX ± 1), a ∈ Z, which is impossible. Thus f is irreducible as well.] We have D(g) = −4(−7)3 − 27(−2)2 = 1264, which is not a square in Q. Thus m = 6, so the Galois group of f is S4 .

page 1 of Chapter 7

CHAPTER 7 INTRODUCING ALGEBRAIC NUMBER THEORY (COMMUTATIVE ALGEBRA 1) The general theory of commutative rings is known as commutative algebra. The main applications of this discipline are to algebraic number theory, to be discussed in this chapter, and algebraic geometry, to be introduced in Chapter 8. Techniques of abstract algebra have been applied to problems in number theory for a long time, notably in the eﬀort to prove Fermat’s Last Theorem. As an introductory example, we will sketch a problem for which an algebraic approach works very well. If p is an odd prime and p ≡ 1 mod 4, we will prove that p is the sum of two squares, that is, p can be expressed as x2 + y 2 where x and y are integers. Since p−1 2 is even, it follows that -1 is a quadratic residue (that is, a square) mod p. [Pair each of the numbers 2,3,. . . ,p − 2 with its inverse mod p and pair 1 with p − 1 ≡ −1 mod p. The product of the numbers 1 through p − 1 is, mod p, 1 × 2 × ··· ×

p−1 p−1 × −1 × −2 × · · · × − 2 2

2 and therefore [( p−1 2 )!] ≡ −1 mod p.]

If −1 ≡ x2 mod p, then p divides x2 + 1. Now we enter the ring of Gaussian integers and factor x2 + 1 as (x + i)(x − i). Since p can divide neither factor, it follows that p is not prime in Z[i], so we can write p = αβ where neither α nor β is a unit. Deﬁne the norm of γ = a + bi as N (γ) = a2 + b2 . Then N (γ) = 1 iﬀ γ = ±1 or ±i iﬀ γ is a unit. (See Section 2.1, Problem 5.) Thus p2 = N (p) = N (α)N (β) with N (α) > 1 and N (β) > 1, so N (α) = N (β) = p. If α = x + iy, then p = x2 + y 2 . Conversely, if p is an odd prime and p = x2 + y 2 , then p is congruent to 1 mod 4. (If x is even, then x2 ≡ 0 mod 4, and if x is odd, then x2 ≡ 1 mod 4. We cannot have x and y both even or both odd, since p is odd.) It is natural to conjecture that we can identify those primes that √ can be represented as x2 + |d|y 2 , where d is a negative integer, by working in the ring Z[ d]. But the Gaussian integers (d = −1) form a Euclidean domain, in particular a unique factorization domain. On the other hand, unique factorization fails for d ≤ −3 (Section 2.7, Problem 7), so the above argument collapses. [Recall from (2.6.4) that in a UFD, an element p that is not prime must be reducible.] Diﬃculties of this sort led Kummer to invent “ideal numbers”, which later became ideals at the hands of Dedekind. We will see that although a ring of algebraic integers need not be a UFD, unique factorization of ideals will always hold. 7.1 Integral Extensions If E/F is a ﬁeld extension and α ∈ E, then α is algebraic over F iﬀ α is a root of a polynomial with coeﬃcients in F . We can assume if we like that the polynomial is monic, and this turns out to be crucial in generalizing the idea to ring extensions. 7.1.1 Deﬁnitions and Comments In this chapter, unless otherwise speciﬁed, all rings are assumed commutative. Let A be a subring of the ring R, and let x ∈ R. We say that x is integral over A if x is a root of a monic polynomial f with coeﬃcients in A. The equation f (X) = 0 is called an equation of integral dependence for x over A. If x is a real or complex number that is integral over Z, then x is called an algebraic integer. Thus for every √ integer d, d is an algebraic integer, as is any nth root of unity. (The monic polynomials are, respectively, X 2 − d and X n − 1.) In preparation for the next result on conditions

page 2 of Chapter 7

equivalent to integrality, note that A[x], the set of polynomials in x with coeﬃcients in A, is an A-module. (The sum of two polynomials is a polynomial, and multiplying a polynomial by a member of A produces another polynomial over A.) 7.1.2 Proposition Let A be a subring of R, with x ∈ R. The following conditions are equivalent: (i) x is integral over A; (ii) The A-module A[x] is ﬁnitely generated; (iii) x belongs to a subring B of R such that A ⊆ B and B is a ﬁnitely generated A-module. Proof. (i) implies (ii). If x is a root of a monic polynomial over A of degree n, then xn and all higher powers of x can be expressed as linear combinations of lower powers of x. Thus 1, x, x2 , , . . . , xn−1 generate A[x] over A. (ii) implies (iii). Take B = A[x]. (iii) implies(i). If β1 , . . . , βn generate B over A, then xβi is a linear combination of the βj , n say xβi = j=1 cij βj . Thus if β is a column vector whose components are the βi , I is an n by n identity matrix, and C = [cij ], then (xI − C)β = 0, and if we premultiply by the adjoint matrix of xI − C (as in Cramer’s rule), we get [det(xI − C)]Iβ = 0, hence [det(xI − C)]b = 0 for every b ∈ B. Since B is a ring we may set b = 1 and conclude that x is a root of the monic polynomial det(XI − C) in A[X]. ♣ For other equivalent conditions, see Problems 1 and 2. We are going to prove a transitivity property for integral extensions (analogous to (3.3.5)), and the following result will be helpful. 7.1.3 Lemma Let A be a subring of R, with x1 , . . . , xn ∈ R. If x1 is integral over A, x2 is integral over A[x1 ],. . . , and xn is integral over A[x1 , . . . , xn−1 ], then A[x1 , . . . , xn ] is a ﬁnitely generated A-module. Proof. The n = 1 case follows from (7.1.2), part (ii). Going from n − 1 to n amounts to proving that if A, B and C are rings, with C a ﬁnitely generated B-module and B a ﬁnitely generated A-module, then C is a ﬁnitely generated A-module. This follows by a brief computation: C=

r j=1

Byj , B =

s k=1

Azk so C =

r s

Ayj zk . ♣

j=1 k=1

7.1.4 Transitivity of Integral Extensions Let A, B and C be subrings of R. If C is integral over B, that is, each element of C is integral over B, and B is integral over A, then C is integral over A. Proof. Let x ∈ C, with xn + bn−1 xn−1 + · · · + b1 x + b0 = 0, bi ∈ B. Then x is integral over A[b0 , . . . , bn−1 ]. Each bi is integral over A, hence over A[b0 , . . . , bi−1 ]. By (7.1.3), A[b0 , . . . , bn , x] is a ﬁnitely generated A-module. By (7.1.2), part (iii), x is integral over A. ♣ 7.1.5 Deﬁnitions and Comments If A is a subring of R, the integral closure of A in R is the set Ac of elements of R that are integral over A. Note that A ⊆ Ac because each

page 3 of Chapter 7

a ∈ A is a root of X − a. We say that A is integrally closed in R if Ac = A. If we simply say that A is integrally closed without reference to R, we assume that A is an integral domain with quotient ﬁeld K, and A is integrally closed in K. If x and y are integral over A, then just as in the proof of (7.1.4), it follows from (7.1.3) that A[x, y] is a ﬁnitely generated A-module. Since x+y, x−y and xy belong to this module, they are integral over A by (7.1.2) part (iii). The important conclusion is that Ac is a subring of R containing A. If we take the integral closure of the integral closure, we get nothing new. 7.1.6 Proposition The integral closure Ac of A in R is integrally closed in R. Proof. By deﬁnition, Ac is integral over A. If x is integral over Ac , then as in the proof of (7.1.4), x is integral over A, so that x ∈ Ac . ♣ We can identify a large class of integrally closed rings. 7.1.7 Proposition If A is a UFD, then A is integrally closed. Proof. If x belongs to the quotient ﬁeld K, then we can write x = a/b where a, b ∈ A, with a and b relatively prime. If x is integral over A, then there is an equation of the form (a/b)n + an−1 (a/b)n−1 + · · · + a1 (a/b) + a0 = 0 with ai ∈ A. Multiplying by bn , we have an + bc = 0, with c ∈ A. Thus b divides an , which cannot happen for relatively prime a and b unless b has no prime factors at all, in other words, b is a unit. But then x = ab−1 ∈ A. ♣ We can now discuss one of the standard setups for doing algebraic number theory. 7.1.8 Deﬁnitions and Comments A number ﬁeld is a subﬁeld L of the complex numbers C such that L is a ﬁnite extension of the rationals Q. Thus the elements of L are algebraic numbers. The integral closure of Z in L is called the ring of algebraic integers (or simply integers) of L. In the next section, we will ﬁnd the algebraic integers explicitly when L is a quadratic extension. Problems For Section 7.1 1. Show that in (7.1.2) another equivalent condition is the following: (iv) There is a subring B of R such that B is a ﬁnitely generated A-module and xB ⊆ B. If R is a ﬁeld, show that the assumption that B is a subring can be dropped (as long as B = 0). 2. A module is said to be faithful if its annihilator is 0. Show that in (7.1.2) the following is another equivalent condition: (v) There is a faithful A[x]-module B that is ﬁnitely generated as an A-module. Let A be a subring of the integral domain B, with B integral over A. In Problems 3-5 we are going to show that A is a ﬁeld if and only if B is a ﬁeld. 3. Assume that B is a ﬁeld, and let a be a nonzero element of A. Then since a−1 ∈ B, there is an equation of the form (a−1 )n + cn−1 (a−1 )n−1 + · · · + c1 a−1 + c0 = 0 with ci ∈ A. Show that a−1 ∈ A, proving that A is a ﬁeld. 4. Now assume that A is a ﬁeld, and let b be a nonzero element of B. By (7.1.2) part (ii), A[b] is a ﬁnite-dimensional vector space over A. Let f be the A-linear transformation on this vector space given by multiplication by b, in other words, f (z) = bz, z ∈ A[b]. Show

page 4 of Chapter 7

that f is injective. 5. Show that f is surjective as well, and conclude that B is a ﬁeld. In Problems 6-8, let A be a subring of B, with B integral over A. Let Q be a prime ideal of B and let P = Q ∩ A. 6. Show that P is a prime ideal of A, and that A/P can be regarded as a subring of B/Q. 7. Show that B/Q is integral over A/P . 8. Show that P is a maximal ideal of A if and only if Q is a maximal ideal of B. 7.2 Quadratic Extensions of the Rationals

√ We will determine the algebraic integers of L = Q( d), where d is a square-free integer (a product of primes). The restriction on d involves no loss of generality; for √ √ distinct √ example, Q( 12) = Q( 3). The minimal polynomial of d over Q is X 2 − d, which has √ roots ± d. The extension√ L/Q is Galois, √ and the Galois group consists of the identity and the automorphism σ(a + b d) = a − b d, a, b ∈ Q. A remark on notation: To make sure that there is no confusion between algebraic integers and ordinary integers, we will use the term rational integer for a member of Z. √ 7.2.1 Lemma If a and b are rational numbers, then a + b d is an algebraic integer if and only if 2a and a2 − db2 belong to Z. In this case, 2b is also in Z. √ √ Proof. Let x = a + b d, so that σ(x) = a − b d. Then x + σ(x) = 2a ∈ Q and xσ(x) = a2 − db2 ∈ Q. Now if x is an algebraic integer, then x is a root of a monic polynomial f ∈ Z[X]. But f (σ(x)) = σ(f (x)) since σ is an automorphism, so σ(x) is also a root of f and hence an algebraic integer. By (7.1.5), 2a and a2 − db2 are also algebraic integers, as well as rational numbers. By (7.1.7), Z is integrally closed, so 2a and a2 − db2 belong to Z. √ The converse holds because a+b d is a root of (X −a)2 = db2 , i.e., X 2 −2aX +a2 −db2 = 0.

Now if 2a and a2 − db2 are rational integers, then (2a)2 − d(2b)2 = 4(a2 − db2 ) ∈ Z, so d(2b)2 ∈ Z. If 2b ∈ / Z, then its denominator would include a prime factor p , which would appear as p2 in the denominator of (2b)2 . Multiplication of (2b)2 by d cannot cancel the p2 because d is square-free, and the result follows. ♣ √ 7.2.2 Corollary The set B of algebraic integers of Q( d), d square-free, can be described as follows. √ (i) If d ≡ 1 mod 4, then B consists of all a + b d, a, b ∈ Z; √ (ii) If d ≡ 1 mod 4, then B consists of all u2 + v2 d, u, v ∈ Z, where u and v have the same parity (both even or both odd). [Note that since d is square-free, it is not divisible by 4, so the condition in (i) is d ≡ 2 or 3 mod 4.] √ Proof. By (7.2.1), the algebraic integers are of the form u2 + v2 d where u, v ∈ Z and u2 dv 2 2 2 4 − 4 ∈ Z, i.e., u − dv ≡ 0 mod 4. It follows that u and v have the same parity. (The square of an even number is congruent to 0 and the square of an odd number to 1 mod 4.) Moreover, the “both odd” case can only occur when d ≡ 1 mod 4. The “both even” case is equivalent to u2 , v2 ∈ Z, and the result follows. ♣ We can express these results in a more convenient form. We will show in (7.4.10) that the set B of algebraic integers in any number ﬁeld L is a free Z-module of rank n = [L : Q]. A basis for this module is called an integral basis or Z-basis for B. √ 7.2.3 Theorem Let B be the algebraic integers of Q( d), d square-free. √ (i) If d ≡ 1 mod 4, then 1 and d form an integral basis of B; √ (ii) If d ≡ 1 mod 4, then 1 and 12 (1 + d) form an integral basis.

page 5 of Chapter 7

Proof. (i) By (7.2.2), 1 and irrational.

√

d span B over Z, and they are linearly independent because

√

d is

√ (ii) By (7.2.2), 1 and 12 (1 + d) are algebraic integers. To show that they span B, consider √ 1 2 (u + v d), where u and v have the same parity. Then √ √ 1 u−v 1 (u + v d) = ( )(1) + v[ (1 + d)] 2 2 2

with

u−v 2

and v in Z. Finally, to show linear independence, assume that a, b ∈ Z and √ 1 a + b[ (1 + d)] = 0. 2

√ Then 2a + b + b d = 0, which forces a = b = 0. ♣ Problems For Section 7.2 2 1. Let L = Q(α), where √ α is a root of the irreducible quadratic X + bX + c, with b, c ∈ Q. Show that L = Q( d) for some square-free integer d. Thus the analysis of this section covers all possible quadratic extensions of√Q. 2. Show that the quadratic extensions Q( d), d square-free, are all distinct. 3. Continuing Problem 2, show that in fact no two distinct quadratic extensions of Q are Q-isomorphic.

Cyclotomic ﬁelds do not exhibit the same behavior. Let ωn = ei2π/n , a primitive 2 = ωn , and n root of unity. By a direct computation, we have ω2n th

n+1 −ω2n = −eiπ(n+1)/n = eiπ eiπ eiπ/n = ω2n .

4. Show that if n is odd, then Q(ωn ) = Q(ω2n ). 5. If x is an algebraic integer, show that the minimal polynomial of x over Q has coeﬃcients in Z. (This will be a consequence of the general theory to be developed in this chapter, but it is accessible now without heavy machinery.) Consequently, an algebraic integer that belongs to Q in fact belongs to Z. (The minimal polynomial of r ∈ Q over Q is X − r.) 6. Give an example of a quadratic extension of Q that is also a cyclotomic extension. 7. Show that an integral basis for the ring of algebraic integers of a number ﬁeld L is, in particular, a basis for L over Q. 7.3 Norms and Traces 7.3.1 Deﬁnitions and Comments If E/F is a ﬁeld extension of ﬁnite degree n, then in particular, E is an n-dimensional vector space over F , and the machinery of basic linear algebra becomes available. If x is any element of E, we can study the F -linear transformation m(x) given by multiplication by x, that is, m(x)y = xy. We deﬁne the norm and the trace of x, relative to the extension E/F , as N [E/F ](x) = det m(x) and T [E/F ](x) = trace m(x). We will write N (x) and T (x) if E/F is understood. If the matrix A(x) = [aij (x)] represents m(x) with respect to some basis for E over F , then the norm of x is the determinant of A(x) and the trace of x is the trace of A(x), that is, the sum of the main diagonal entries. The characteristic polynomial of x is deﬁned as the characteristic polynomial of the matrix A(x), that is, char[E/F ](x) = det[XI − A(x)]

page 6 of Chapter 7

where I is an n by n identity matrix. If E/F is understood, we will refer to the characteristic polynomial of x, written char(x). 7.3.2 Example Let E = C and F = R. A basis for C over R is {1, i} and, with x = a + bi, we have (a + bi)(1) = a(1) + b(i) and (a + bi)(i) = −b(1) + a(i).

Thus A(a + bi) =

a −b . b a

The norm, trace and characteristic polynomial of a + bi are N (a + bi) = a2 + b2 , T (a + bi) = 2a, char(a + bi) = X 2 − 2aX + a2 + b2 . The computation is exactly the same if E = Q(i) and F = Q. Notice that the coeﬃcient of X is minus the trace and the constant term is the norm. In general, it follows from the deﬁnition of characteristic polynomial that char(x) = X n − T (x)X n−1 + · · · + (−1)n N (x). [The only terms multiplying X n−1 in the expansion of the determinant are −aii (x), i = 1, . . . , n. Set X = 0 to show that the constant term of char(x) is (−1)n det A(x).] 7.3.3 Lemma If E is an extension of F and x ∈ E, then N (x), T (x) and the coeﬃcients of char(x) belong to F . If a ∈ F , then N (a) = an , T (a) = na, and char(a) = (X − a)n . Proof. The ﬁrst assertion follows because the entries of the matrix A(x) are in F . The second statement holds because if a ∈ F , the matrix representing multiplication by a is aI. ♣ It is natural to look for a connection between the characteristic polynomial of x and the minimal polynomial of x over F . 7.3.4 Proposition char[E/F ](x) = [min(x, F )]r where r = [E : F (x)]. Proof. First assume that r = 1, so that E = F (x). By the Cayley-Hamilton theorem, the linear transformation m(x) satisﬁes char(x), and since m(x) is multiplication by x, x itself is a root of char(x). Thus min(x, F ) divides char(x). But both polynomials have degree n, and the result follows. In the general case, let y1 , . . . , ys be a basis for F (x) over F , and let z1 , . . . , zr be a basis for E over F (x). Then the yi zj form a basis for E over F . Let A = A(x)be the matrix representing multiplication by x in the extension F (x)/F , so that xyi = k aki yk , and x(yi zj ) = k aki (yk zj ). Order the basis for E/F as y1 z1 , y2 z1 , . . . , ys z1 ; y1 z2 , y2 z2 , . . . , ys z2 ; . . . ; y1 zr , y2 zr , . . . , ys zr . Then m(x) is represented in E/F as   A 0 ... 0  0 A ... 0   . . ..   .. .. .  0

0

... A

Thus char[E/F ](x) = [det(XI − A)]r , which by the r = 1 case coincides with [min(x, F )]r . ♣

page 7 of Chapter 7

7.3.5 Corollary Let [E : F ] = n, and [F (x) : F ] = d. Let x1 , . . . , xd be the roots of min(x, F ) in a splitting ﬁeld (counting multiplicity). Then N (x) = (

d

n xi d i=1 d

xi )n/d , T (x) =

i=1

and char(x) = [

d

(X − xi )]n/d .

i=1

Proof. The formula for the characteristic polynomial follows from (7.3.4). The norm is (−1)n times the constant term of char(x) (see (7.3.2)), hence is (−1)n (−1)n (

d

xi )n/d .

i=1

Finally, if min(x, F ) = X d + ad−1 X d−1 + · · · + a1 X + a0 , then the coeﬃcient of X n−1 in d [min(x, F )]n/d is nd ad−1 = − nd i=1 xi . Since the trace is the negative of this coeﬃcient [see (7.3.2)], the result follows. ♣ If E is a separable extension of F , there are very useful alternative expressions for the trace and norm. 7.3.6 Proposition Let E/F be a separable extension of degree n, and let σ1 , . . . , σn be the distinct F -monomorphisms of E into an algebraic closure of E, or equally well into a normal extension L of F containing E. Then T [E/F ](x) =

n

σi (x) and N [E/F ](x) =

i=1

n

σi (x).

i=1

Consequently, T (ax + by) = aT (x) + bT (y) and N (xy) = N (x)N (y) for x, y ∈ E, a, b ∈ F . Proof. Each of the d distinct F -embeddings τi of F (x) into L takes x into a unique conjugate xi , and extends to exactly nd = [E : F (x)] F -embeddings of E into L, all of which also take x to xi [see (3.5.1), (3.2.3) and (3.5.2)]. Thus n

n τi (x) = T (x) d i=1 d

σi (x) =

i=1

and

n i=1

σi (x) = [

d

τi (x)]n/d = N (x)

i=1

by (7.3.5). ♣ The linearity of T and the multiplicativity of N hold without any assumption of separability, since in (7.3.1) we have m(ax + by) = am(x) + bm(y) and m(xy) = m(x) ◦ m(y). 7.3.7 Corollary (Transitivity of Trace and Norm) If F ≤ K ≤ E, where E/F is ﬁnite and separable, then T [E/F ] = T [K/F ] ◦ T [E/K] and N [E/F ] = N [K/F ] ◦ N [E/K]. Proof. Let σ1 , . . . , σn be the distinct F -embeddings of K into L, and let τ1 , . . . , τm be the distinct K-embeddings of E into L, where L is the normal closure of E over F . By (6.3.1)

page 8 of Chapter 7

and (3.5.11), L/F is Galois, and by (3.5.2), (3.5.5) and (3.5.6), each mapping σi and τj extends to an automorphism of L. Therefore it makes sense to allow the mappings to be composed. By (7.3.6), T [K, F ](T [E/K])(x) =

n i=1

m n m σi ( τj (x)) = σi τj (x). j=1

i=1 j=1

Now each σi τj is an F -embedding of E into L, and the number of mappings is mn = [E : K][K : F ] = [E : F ]. Furthermore, the σi τj are distinct when restricted to E. For if σi τj = σk τl on E, hence on K, then σi = σk on K (because τj = τl = the identity on K). Thus i = k, so that τj = τl on E. But then j = l. By (7.3.6), T [K, F ](T [E/K])(x) = T [E/F ](x). The norm is handled the same way, with sums replaced by products. ♣ 7.3.8 Corollary If E/F is a ﬁnite separable extension, then T [E/F ](x) cannot be 0 for all x ∈ E. n Proof. If T (x) = 0 for all x, then by (7.3.6), i=1 σi (x) = 0 for all x. This contradicts Dedekind’s lemma (6.1.6). ♣ A statement equivalent to (7.3.8) is that if E/F is ﬁnite and separable, then the “trace form” [the bilinear form (x, y) → T [E/F ](xy)] is nondegenerate, i.e., if T (xy) = 0 for all y, then x = 0. For if x = 0, T (x0 ) = 0, and T (xy) = 0 for all y, choose y so that xy = x0 to reach a contradiction. 7.3.9 The Basic Setup For Algebraic Number Theory Let A be an integral domain with quotient ﬁeld K, and let L be a ﬁnite separable extension of K. Let B be the set of elements of L that are integral over A, that is, B is the integral closure of A in L. The diagram below summarizes all the information. L | K

B | A

In the most important special case, A = Z, K = Q, L is a number ﬁeld, and B is the ring of algebraic integers of L. From now on, we will refer to (7.3.9) as the AKLB setup. 7.3.10 Proposition If x ∈ B, then the coeﬃcients of char[L/K](x) and min(x, K) are integral over A. In particular, T [L/K](x) and N [L/K](x) are integral over A, by (7.3.2). If A is integrally closed, then by (7.3.3), the coeﬃcients belong to A. Proof. The coeﬃcients of min(x, K) are sums of products of the roots xi , so by (7.1.5) and (7.3.4), it suﬃces to show that the xi are integral over A. Each xi is a conjugate of x over K, so by (3.2.3) there is a K-isomorphism τi : K(x) → K(xi ) such that τi (x) = xi . If we apply τi to an equation of integral dependence for x over A, we get an equation of integral dependence for xi over A. ♣ Problems For Section 7.3 √ √ 1. If E = Q( d) and x = a + b d ∈ E, ﬁnd the norm and trace of x. 2. If E = Q(θ) where θ is a root of the irreducible cubic X 3 − 3X + 1, ﬁnd the norm and trace of θ2 . 3. Find the trace of the primitive 6th root of unity ω in the cyclotomic extension Q6 . We will now prove Hilbert’s Theorem 90: If E/F is a cyclic extension with [E : F ] = n and Galois group G = {1, σ, . . . , σ n−1 } generated by σ, and x ∈ E, then (i) N (x) = 1 if and only if there exists y ∈ E such that x = y/σ(y); (ii) T (x) = 0 if and only if there exists z ∈ E such that x = z − σ(z).

page 9 of Chapter 7

4. Prove the “if” parts of (i) and (ii) by direct computation. By Dedekind’s lemma, 1, σ, σ 2 , . . . , σ n−1 are linearly independent over E, so 1 + xσ + xσ(x)σ 2 + · · · + xσ(x) · · · σ n−2 (x)σ n−1 is not identically 0 on E. 5. Use this to prove the “only if” part of (i). By (7.3.8), there is an element u ∈ E whose trace is not 0. Let w = xσ(u) + (x + σ(x))σ 2 (u) + · · · + (x + σ(x) + · · · + σ n−2 (x))σ n−1 (u) hence σ(w) = σ(x)σ 2 (u) + (σ(x) + σ 2 (x))σ 3 (u) + · · · + (σ(x) + σ 2 (x) + · · · + σ n−1 (x))σ n (u) 6. If T (x) = 0, show that w − σ(w) = xT (u). 7. If z = w/T (u), show that z − σ(z) = x, proving the “only if” part of (ii). 8. In Hilbert’s Theorem 90, are the elements y and z unique? √ 2 3 9. Let θ be a root of X 4 − 2 over Q. Find √ the trace over Q of θ, θ , θ and 3θ. 10. Continuing Problem 9, show that 3 cannot belong to Q[θ]. 7.4 The Discriminant We have met the discriminant of a polynomial in connection with Galois theory (Section 6.6). There is also a discriminant in algebraic number theory. The two concepts are unrelated at ﬁrst glance, but there is a connection between them. We assume the basic AKLB setup of (7.3.9), with n = [L : K]. 7.4.1 Deﬁnition The discriminant of the n-tuple x = (x1 , . . . , xn ) of elements of L is D(x) = det(T [L/K](xi xj )). Thus we form a matrix whose ij element is the trace of xi xj , and take the determinant of the matrix. By (7.3.3) and (7.3.10), D(x) belongs to K and is integral over A, hence belongs to A if A is integrally closed. The discriminant behaves quite reasonably under linear transformation: 7.4.2 Lemma If y = Cx, where C is an n by n matrix and x and y are n-tuples written as column vectors, then D(y) = (det C)2 D(x). Proof. The trace of yr ys is T( cri csj xi xj ) = cri T (xi xj )csj i,j

hence

i,j

(T (yr ys )) = C(T (xi xj ))C

where C is the transpose of C. The result follows upon taking determinants. ♣ Here is an alternative expression for the discriminant. 7.4.3 Lemma Let σ1 , . . . , σn be the K-embeddings of L into an algebraic closure of L, as in (7.3.6). Then D(x) = [det(σi (xj ))]2 .

page 10 of Chapter 7

Thus we form the matrix whose ij element is σi (xj ), take the determinant and square the result. Proof. By (7.3.6), T (xi xj ) =

σk (xi xj ) =

k

σk (xi )σk (xj )

k

so if C is the matrix whose ij entry is σi (xj ), then (T (xi xj )) = C C and again the result follows upon taking determinants. ♣ The discriminant “discriminates” between bases and non-bases, as follows. 7.4.4 Proposition If x = (x1 , . . . , xn ), then the xi form a basis for L over K if and only if D(x) = 0. Proof. If j cj xj = 0, with the cj ∈ K and not all 0, then j cj σi (xj ) = 0 for all i, so the columns of the matrix B = (σi (xj )) are linearly dependent. Thus linear dependence of the xi implies that D = 0. Conversely, assume that the xi are linearly independent (and therefore a basis since n = [L : K]). If D = 0, then the rows of B are linearly dependent, so for some c ∈ K, not all 0, we have i ci σi (xj ) = 0 for all j. Since the xj form a basis, we i have i ci σi (u) = 0 for all u ∈ L, so the monomorphisms σi are linearly dependent. This contradicts Dedekind’s lemma. ♣ We now make the connection between the discriminant deﬁned above and the discriminant of a polynomial deﬁned previously. 7.4.5 Proposition Assume that L = K(x), and let f be the minimal polynomial of x over K. Let D be the discriminant of the basis 1, x, x2 , . . . , xn−1 for L over K. Then D is the discriminant of the polynomial f . Proof. Let x1 , . . . , xn be the roots of f in a splitting ﬁeld, with x1 = x. Let σi be the K-embedding that takes x to xi , i = 1, . . . , n. Then σi (xj ) = xji , so by (7.4.3), D is the square of the determinant of the matrix 1 1 . . . 1

x1 x2 .. .

x21 x22

xn

x2n

 · · · xn−1 1 n−1 · · · x2   ..  . . . . xn−1 n

and the result follows from the formula for a Vandermonde determinant; see (A6.2). ♣ 7.4.6 Corollary Under the hypothesis of (7.4.5), n D = (−1)( 2 ) N [L/K](f (x))

where f is the derivative of f . n Proof. Let a = (−1)( 2 ) . By (7.4.5),

D=

(xi − xj )2 = a

i<j

(xi − xj ) = a (xi − xj ). i j=i

i=j

But f (X) = (X − x1 ) · · · (X − xn ), so f (xi ) =

k j=k

(X − xj )

page 11 of Chapter 7

with X replaced by xi . When X is replaced by xi , only the k = i term is nonzero, hence (xi − xj ).

f (xi ) =

j=i

Consequently, D=a

n

f (xi ).

i=1

But

f (xi ) = f (σi (x)) = σi (f (x))

so by (7.3.6),

D = aN [L/K](f (x)). ♣

The discriminant of an integral basis for a number ﬁeld has special properties. We will get at these results by considering the general AKLB setup, adding some additional conditions as we go along. 7.4.7 Lemma There is a basis for L/K consisting entirely of elements of B. Proof. Let x1 , . . . , xn be a basis for L over K. Each xi is algebraic over K, and therefore satisﬁes a polynomial equation of the form am xm i + · · · + a1 xi + a0 = 0 with am = 0 and the ai ∈ A. (Initially, we only have ai ∈ K, but then ai is the ratio of two elements in A, and we can form a common denominator.) Multiply the equation by am−1 m to obtain an equation of integral dependence for yi = am xi over A. The yi form the desired basis. ♣ 7.4.8 Theorem Suppose we have a nondegenerate symmetric bilinear form on an ndimensional vector space V , written for convenience using inner product notation (x, y). If x1 , . . . , xn is any basis for V , then there is a basis y1 , . . . , yn for V , called the dual basis referred to V , such that

1, i = j; (xi , yj ) = δij = 0, i = j. This is a standard (and quite instructive) result in linear algebra, and it will be developed in the exercises. 7.4.9 Theorem If A is a principal ideal domain, then B is a free A-module of rank n. Proof. By (7.3.8), the trace is a nondegenerate symmetric bilinear form on the n-dimensional vector space L over K. By (7.1.7), A is integrally closed, so by (7.3.10), the trace of any element of L belongs to A. Now let x1 , . . . , xn be any basis for L over K, and let y1 , . . . , yn n be the dual basis referred to L (see (7.4.8)). If z ∈ B, then we can write z = j=1 aj yj with aj ∈ K. We know that the trace of xi z belongs to A, and we also have n n n T (xi z) = T ( aj xi yj ) = aj T (xi yj ) = aj δij = ai . j=1

j=1

j=1

Thus each ai belongs to A, so that B is an A-submodule of the free A-module ⊕nj=1 Ayj . By (4.6.2), B is a free A-module of rank at most n. But by (7.4.7), B contains a basis for L over K, and if we wish, we can assume that this basis is x1 , . . . , xn . Then B contains the free A-module ⊕nj=1 Axj , so the rank of B as an A-module is at least n, and hence exactly n. ♣

page 12 of Chapter 7

7.4.10 Corollary The set B of algebraic integers in any number ﬁeld L is a free Z-module of rank n = [L : Q]. Therefore B has an integral basis. The discriminant is the same for every integral basis; it is known as the ﬁeld discriminant. Proof. Take A = Z in (7.4.9) to show that B has an integral basis. The transformation matrix C between two integral bases (see (7.4.2)) is invertible, and both C and C −1 have rational integer coeﬃcients. Take determinants in the equation CC −1 = I to conclude that det C is a unit in Z. Therefore det C = ±1, so by (7.4.2), all integral bases have the same discriminant. ♣ Problems For Section 7.4 Let x1 , . . . , xn be a basis for the vector space V , and let (x, y) be a nondegenerate symmetric bilinear form on V . We now supply the details of the proof of (7.4.8). 1. For any y ∈ V , the mapping x → (x, y) is a linear form l(y), i.e., a linear map from V to the ﬁeld of scalars. Show that the linear transformation y → l(y) from V to V ∗ , the dual space of V (i.e., the space of all linear forms on V ), is injective. 2. Show that any linear form on V is l(y) for some y ∈ V . 3. Let f1 , . . . , fn be the dual basis corresponding to x1 , . . . , xn . Thus each fj belongs to V ∗ (not V ) and fj (xi ) = δij . If fj = l(yj ), show that y1 , . . . , yn is the required dual basis referred to V . n 4. Show that xi = j=1 (xi , xj )yj . Thus in order to compute the dual basis referred to V in terms of the original basis, we must invert the matrix ((xi , xj )). 5. A matrix C with coeﬃcients in Z is said to be unimodular if its determinant is ±1. Show that C is unimodular if and only if C is invertible and its inverse √ has coeﬃcients in Z. 6. Show that the ﬁeld discriminant of the quadratic extension Q( d), d square-free, is D=

4d d

if d ≡ 1 mod 4; if d ≡ 1 mod 4.

7. Let x1 , . . . , xn be arbitrary algebraic integers in a number ﬁeld, and consider the determinant of the matrix (σi (xj )), as in (7.4.3). The direct expansion of the determinant has n! terms. Let P be the sum of those terms in the expansion that have plus signs in front of them, and N the sum of those terms preﬁxed by minus signs. Thus the discriminant D of (x1 , . . . , xn ) is (P − N )2 . Show that P + N and P N are ﬁxed by each σi , and deduce that P + N and P N are rational numbers. 8. Continuing Problem 7, show that P + N and P N are rational integers. 9. Continuing Problem 8, prove Stickelberger’s theorem: D ≡ 0 or 1 mod 4. 10. Let L be a number ﬁeld of degree n over Q, and let y1 , . . . , yn be a basis for L over Q consisting of algebraic integers. Let x1 , . . . , xn be an integral basis. Show that if the discriminant D(y1 , . . . , yn ) is square-free, then each xi can be expressed as a linear combination of the yj with integer coeﬃcients. 11. Continuing Problem 10, show that if D(y1 , . . . , yn ) is square-free, then y1 , . . . , yn is an integral basis. 12. Is the converse of the result of Problem 11 true? 13. In the standard AKLB setup (see (7.3.9)), show that L is the quotient ﬁeld of B. 7.5 Noetherian and Artinian Modules and Rings 7.5.1 Deﬁnitions and Comments In this section, rings are not assumed commutative. Let M be an R-module, and suppose that we have an increasing sequence of submodules M1 ≤ M2 ≤ M3 ≤ . . ., or a decreasing sequence M1 ≥ M2 ≥ M3 ≥ . . .. We say that the sequence stabilizes if for some t, Mt = Mt+1 = Mt+2 = . . .. The question of stabilization of sequences of submodules appears in a fundamental way in many areas of abstract algebra and its applications. We have already made contact with the idea; see (2.6.6) and the introductory remarks in Section 4.6.

page 13 of Chapter 7

The module M is said to satisfy the ascending chain condition (acc) if every increasing sequence of submodules stabilizes; M satisﬁes the descending chain condition (dcc) if every decreasing sequence of submodules stabilizes. 7.5.2 Proposition The following conditions on an R-module M are equivalent, and deﬁne a Noetherian module: (1) M satisﬁes the acc; (2) Every nonempty collection of submodules of M has a maximal element (with respect to inclusion). The following conditions on M are equivalent, and deﬁne an Artinian module:

(1 ) M satisﬁes the dcc; (2 ) Every nonempty collection of submodules of M has a minimal element. Proof. Assume (1), and let S be a nonempty collection of submodules. Choose M1 ∈ S. If M1 is maximal, we are ﬁnished; otherwise we have M1 < M2 for some M2 ∈ S. If we continue inductively, the process must terminate at a maximal element; otherwise the acc would be violated. Conversely, assume (2), and let M1 ≤ M2 ≤ . . .. The sequence must stabilize; otherwise {M1 , M2 , . . .} would be a nonempty collection of submodules with no maximal element. The proof is exactly the same in the Artinian case, with all inequalities reversed. ♣ There is another equivalent condition in the Noetherian case. 7.5.3 Proposition M is Noetherian iﬀ every submodule of M is ﬁnitely generated. Proof. If the sequence M1 ≤ M2 ≤ . . . does not stabilize, let N = ∪∞ r=1 Mr . Then N is a submodule of M , and it cannot be ﬁnitely generated. For if x1 , . . . , xs generate N , then for suﬃciently large t, all the xi belong to Mt . But then N ⊆ Mt ⊆ Mt+1 ⊆ · · · ⊆ N , so Mt = Mt+1 = . . .. Conversely, assume that the acc holds, and let N ≤ M . If N = 0, choose x1 ∈ N . If Rx1 = N , then N is ﬁnitely generated. Otherwise, there exists x2 ∈ / Rx1 . If / Rx1 + Rx2 . The acc x1 and x2 generate N , we are ﬁnished. Otherwise, there exists x3 ∈ forces the process to terminate at some stage t, in which case x1 , . . . , xt generate N . ♣ The analogous equivalent condition in the Artinian case (see Problem 8) is that every quotient module M/N is ﬁnitely cogenerated, that is, if the intersection of a collection of submodules of M/N is 0, then there is a ﬁnite subcollection whose intersection is 0. 7.5.4 Deﬁnitions and Comments A ring R is Noetherian [resp. Artinian] if it is Noetherian [resp. Artinian] as a module over itself. If we need to distinguish between R as a left, as opposed to right, R-module, we will refer to a left Noetherian and a right Noetherian ring, and similarly for Artinian rings. This problem will not arise until Chapter 9. 7.5.5 Examples 1. Every PID is Noetherian. This follows from (7.5.3), since every ideal is generated by a single element. 2. Z is Noetherian (a special case of Example 1) but not Artinian. There are many descending chains of ideals that do not stabilize, e.g., Z ⊃ (2) ⊃ (4) ⊃ (8) ⊃ . . . . We will prove in Chapter 9 that an Artinian ring must also be Noetherian. 3. If F is a ﬁeld, then the polynomial ring F [X] is Noetherian (another special case of Example 1) but not Artinian. A descending chain of ideals that does not stabilize is (X) ⊃ (X 2 ) ⊃ (X 3 ) ⊃ . . . .

page 14 of Chapter 7

4. The ring F [X1 , X2 , . . .] of polynomials over F in inﬁnitely many variables is neither Artinian nor Noetherian. A descending chain of ideals that does not stabilize is constructed as in Example 3, and an ascending chain of ideals that does not stabilize is (X1 ) ⊂ (X1 , X2 ) ⊂ (X1 , X2 , X3 ) ⊂ . . . . 7.5.6 Remark The following observations will be useful in deriving properties of Noetherian and Artinian modules. If N ≤ M , then a submodule L of M that contains N can always be written in the form K + N for some submodule K. (K = L is one possibility.) By the correspondence theorem, (K1 + N )/N = (K2 + N )/N implies K1 + N = K2 + N and (K1 + N )/N ≤ (K2 + N )/N implies K1 + N ≤ K2 + N. 7.5.7 Proposition If N is a submodule of M , then M is Noetherian [resp. Artinian] if and only if N and M/N are Noetherian [resp. Artinian]. Proof. Assume M is Noetherian. Then N is Noetherian by (2) of (7.5.2), since a submodule of N must also be a submodule of M . By (7.5.6), an ascending chain of submodules of M/N looks like (M1 + N )/N ≤ (M2 + N )/N ≤ · · ·. But then the Mi + N form an ascending sequence of submodules of M , which must stabilize. Consequently, the sequence (Mi + N )/N, i = 1, 2, . . ., must stabilize. Conversely, assume that N and M/N are Noetherian, and let M1 ≤ M2 ≤ · · · be an increasing sequence of submodules of M . Take i large enough so that both sequences {Mi ∩ N } and {Mi + N } have stabilized. If x ∈ Mi+1 , then x + N ∈ Mi+1 + N = Mi + N , so x = y + z where y ∈ Mi and z ∈ N . Thus x − y ∈ Mi+1 ∩ N = Mi ∩ N , and since y ∈ Mi we have x ∈ Mi as well. Consequently, Mi = Mi+1 and the sequence of Mi ’s has stabilized. The Artinian case is handled by reversing inequalities (and interchanging indices i and i + 1 in the second half of the proof). ♣ 7.5.8 Corollary If M1 , . . . , Mn are Noetherian [resp. Artinian] R-modules, then so is M1 ⊕ M 2 ⊕ · · · ⊕ M n . Proof. It suﬃces to consider n = 2 (induction will take care of higher values of n). The submodule N = M1 of M = M1 ⊕ M2 is Noetherian by hypothesis, and M/N ∼ = M2 is also Noetherian (apply the ﬁrst isomorphism theorem to the natural projection of M onto M2 ). By (7.5.7), M is Noetherian. The Artinian case is done the same way. ♣ 7.5.9 Corollary If M is a ﬁnitely generated module over the Noetherian [resp. Artinian] ring R, then M is Noetherian [resp. Artinian] Proof. By (4.3.6), M is a quotient of a free module L of ﬁnite rank. Since L is the direct sum of a ﬁnite number of copies of R, the result follows from (7.5.8) and (7.5.7). ♣ Ascending and descending chains of submodules are reminiscent of normal and subnormal series in group theory, and in fact we can make a precise connection. 7.5.10 Deﬁnitions A series of length n for a module M is a sequence of the form M = M0 ≥ M1 ≥ · · · ≥ Mn = 0. The series is called a composition series if each factor module Mi /Mi+1 is simple. [A module is simple if it is nonzero and has no submodules except itself and 0. We will study simple modules in detail in Chapter 9.] Thus we are requiring the series to have no proper reﬁnement. Two series are equivalent if they have the same length and the same factor modules, up to isomorphism and rearrangement.

page 15 of Chapter 7

By convention, the zero module has a composition series, namely {0} itself. 7.5.11 Jordan-H¨ older Theorem For Modules If M has a composition series, then any two composition series for M are equivalent. Furthermore, any strictly decreasing sequence of submodules can be reﬁned to a composition series. Proof. The development of the Jordan-H¨ older theorem for groups can be taken over verbatim if we change multiplicative to additive notation. In particular, we can reproduce the preliminary lemma (5.6.2), the Zassenhaus lemma (5.6.3), the Schreier reﬁnement theorem (5.6.5), and the Jordan-H¨ older Theorem (5.6.6). We need not worry about normality of subgroups because in an abelian group, all subgroups are normal. As an example of the change in notation, the Zassenhaus lemma becomes A + (B ∩ D) ∼ C + (D ∩ B) . = A + (B ∩ C) C + (D ∩ A) This type of proof can be irritating, because it forces readers to look at the earlier development and make sure that everything does carry over. A possible question is “Why can’t a composition series S of length n coexist with an inﬁnite ascending or descending chain?” But if such a situation occurs, we can form a series T for M of length n + 1. By Schreier, S and T have equivalent reﬁnements. Since S has no proper reﬁnements, and equivalent reﬁnement have the same length, we have n ≥ n + 1, a contradiction. ♣ We can now relate the ascending and descending chain conditions to composition series. 7.5.12 Theorem The R-module M has a composition series if and only if M is both Noetherian and Artinian. Proof. The “only if” part was just done at the end of the proof of (7.5.11). Thus assume that M is Noetherian and Artinian. Assuming (without loss of generality) that M = 0, it follows from (2) of (7.5.2) that M0 = M has a maximal proper submodule M1 . Now M1 is Noetherian by (7.5.7), so if M1 = 0, then M1 has a maximal proper submodule M2 . Continuing inductively, we must reach 0 at some point because M is Artinian. By construction, each Mi /Mi+1 is simple, and we have a composition series for M . ♣ Here is a connection with algebraic number theory. 7.5.13 Proposition In the basic AKLB setup of (7.3.9), assume that A is integrally closed. If A is a Noetherian ring, then so is B. In particular, the ring of algebraic integers in a number ﬁeld is Noetherian. Proof. By the proof of (7.4.9), B is a submodule of a free A-module M of ﬁnite rank. (The assumption that A is a PID in (7.4.9) is used to show that A is integrally closed, and we have this by hypothesis. The PID assumption is also used to show that B is a free A-module, but we do not need this in the present argument.) By (7.5.8), M is Noetherian, so by (7.5.7), B is a Noetherian A-module. An ideal of B is, in particular, an A-submodule of B, hence is ﬁnitely generated over A and therefore over B. Thus B is a Noetherian ring. ♣ Problems For Section 7.5 1. Let p be a ﬁxed prime, and let A be the abelian group of all rational numbers a/pn , n = 0, 1, . . . , a ∈ Z, where all calculations are modulo 1, in other words, A is a subgroup of Q/Z. Let An be the subgroup {0, 1/pn , 2/pn , . . . , (pn − 1)/pn }. Show that A is not a Noetherian Z-module. 2. Continuing Problem 1, if B is a proper subgroup of A, show that B must be one of the An . Thus A is an Artinian Z-module. [This situation cannot arise for rings, where Artinian implies Noetherian.] 3. If V is a vector space, show that V is ﬁnite-dimensional iﬀ V is Noetherian iﬀ V is Artinian iﬀ V has a composition series.

page 16 of Chapter 7

4. Deﬁne the length of a module M [notation l(M )] as the length of a composition series for M . (If M has no composition series, take l(M ) = ∞.) Suppose that we have a short exact sequence f g 0 → N → M → M/N → 0. Show that l(M ) is ﬁnite if and only if l(N ) and l(M/N ) are both ﬁnite. 5. Show that l is additive, that is, l(M ) = l(N ) + l(M/N ). 6. Let S be a subring of the ring R, and assume that S is a Noetherian ring. If R is ﬁnitely generated as a module over S, show that R is also a Noetherian ring. 7. Let R be a ring, and assume that the polynomial ring R[X] is Noetherian. Does it follow that R is Noetherian? 8. Show that a module M is Artinian if and only if every quotient module M/N is ﬁnitely cogenerated. 7.6 Fractional Ideals Our goal is to establish unique factorization of ideals in a Dedekind domain, and to do this we will need to generalize the notion of ideal. First, some preliminaries. 7.6.1 Deﬁnition If I1 , . . . , In are ideals, the product I1 · · · In is the set of all ﬁnite sums a i 1i a2i · · · ani , where aki ∈ Ik , k = 1, . . . , n. It follows from the deﬁnition that the product is an ideal contained in each Ij . 7.6.2 Lemma If P is a prime ideal that contains a product I1 · · · In of ideals, then P contains Ij for some j. Proof. If not, let aj ∈ Ij \ P, j = 1, . . . , n. Then a1 · · · an belongs to I1 · · · In ⊆ P , and since P is prime, some aj belongs to P , a contradiction. ♣ 7.6.3 Proposition If I is a nonzero ideal of the Noetherian integral domain R, then I contains a product of nonzero prime ideals. Proof. Assume the contrary. If S is the collection of all nonzero ideals that do not contain a product of nonzero prime ideals, then since R is Noetherian, S has a maximal element J, and J cannot be prime because it belongs to S. Thus there are elements a, b ∈ R with a∈ / J, b ∈ / J, and ab ∈ J. By maximality of J, the ideals J + Ra and J + Rb each contain a product of nonzero prime ideals, hence so does (J + Ra)(J + Rb) ⊆ J + Rab = J. This is a contradiction. [Notice that we must use the fact that a product of nonzero ideals is nonzero, and this is where the hypothesis that R is an integral domain comes in.] ♣ 7.6.4 Corollary If I is an ideal of the Noetherian ring R (not necessarily an integral domain), then I contains a product of prime ideals. Proof. Repeat the proof of (7.6.3) with the word “nonzero” deleted. ♣ Ideals in the ring of integers are of the form nZ, the set of multiples of n. A set of the form 32 Z is not an ideal because it is not a subset of Z, yet it behaves in a similar manner. The set is closed under addition and multiplication by an integer, and it becomes an ideal of Z if we simply multiply all the elements by 2. It will be proﬁtable to study sets of this type. 7.6.5 Deﬁnitions Let R be an integral domain, with K its quotient ﬁeld, and let I be an R-submodule of K. We say that I is a fractional ideal of R if rI ⊆ R for some nonzero r ∈ R. We will call r a denominator of I. An ordinary ideal of R is a fractional ideal (take r = 1), and will often be referred to an as integral ideal.

page 17 of Chapter 7

7.6.6 Lemma (i) If I is a ﬁnitely generated R-submodule of K, then I is a fractional ideal. (ii) If R is Noetherian and I is a fractional ideal of R, then I is a ﬁnitely generated Rsubmodule of K. (iii) If I and J are fractional ideals with denominators r and s respectively, then I ∩ J, I + J and IJ are fractional ideals with respective denominators r (or s), rs and rs. [The product of fractional ideals is deﬁned exactly as in (7.6.1).] Proof. (i) If x1 = a1 /b1 , . . . , xn = an /bn generate I and b = b1 · · · bn , then bI ⊆ R. (ii) If rI ⊆ R, then I ⊆ r−1 R. As an R-module, r−1 R is isomorphic to R and is therefore Noetherian. Consequently, I is ﬁnitely generated. (iii) It follows from the deﬁnition (7.6.5) that the intersection, sum and product of fractional ideals are fractional ideals. The assertions about denominators are proved by noting that r(I ∩ J) ⊆ rI ⊆ R, rs(I + J) ⊆ rI + sJ ⊆ R, and rsIJ = (rI)(sJ) ⊆ R. ♣ The product of two nonzero fractional ideals is a nonzero fractional ideal, and the multiplication is associative (since multiplication in R is associative). There is an identity element , namely R, since RI ⊆ I = 1I ⊆ RI. We will show that if R is a Dedekind domain, then every nonzero fractional ideal has a multiplicative inverse, so the nonzero fractional ideals form a group. 7.6.7 Deﬁnitions and Comments A Dedekind domain is an integral domain R such that (1) R is Noetherian, (2) R is integrally closed, and (3) Every nonzero prime ideal of R is maximal. Every PID is a Dedekind domain, by (7.5.5), (7.1.7), (2.6.8) and (2.6.9). We will prove that the algebraic integers of a number ﬁeld form a Dedekind domain. But as we know, the ring of algebraic integers need not be a PID, or even a UFD (see the discussion at the beginning of this chapter, and the exercises in Section 7.7). 7.6.8 Lemma Let I be a nonzero prime ideal of the Dedekind domain R, and let J = {x ∈ K : xI ⊆ R}. Then R ⊂ J. Proof. Since RI ⊆ R, it follows that R is a subset of J. Pick a nonzero element a ∈ I, so that I contains the principal ideal Ra. Let n be the smallest positive integer such that Ra contains a product P1 · · · Pn of n nonzero prime ideals. Since R is Noetherian, there is such an n by (7.6.3), and by (7.6.2), I contains one of the Pi , say P1 . But in a Dedekind domain, every nonzero prime ideal is maximal, so I = P1 . Assuming n ≥ 2, set I1 = P2 · · · Pn , so / Ra. Now II1 ⊆ Ra, in particular, that Ra ⊇ I1 by minimality of n. Choose b ∈ I1 with b ∈ Ib ⊆ Ra, hence Iba−1 ⊆ R. (Note that a has an inverse in K, but not necessarily in R.) / R, for if so, b ∈ Ra, contradicting the choice of b. Thus ba−1 ∈ J, but ba−1 ∈ The case n = 1 must be handled separately. In this case, P1 = I ⊇ Ra ⊇ P1 , so I = Ra. Thus Ra is a proper ideal, and we can choose b ∈ R with b ∈ / Ra. Then ba−1 ∈ / R, but −1 −1 −1 ba I = ba Ra = bR ⊆ R, so ba ∈ J. ♣ We now prove that in (7.6.8), J is the inverse of I. 7.6.9 Proposition Let I be a nonzero prime ideal of the Dedekind domain R, and let J = {x ∈ K : xI ⊆ R}. Then J is a fractional ideal and IJ = R. Proof. By deﬁnition, J is an R-submodule of K. If r is a nonzero element of I and x ∈ J, then rx ∈ R, so rJ ⊆ R and J is a fractional ideal. Now IJ ⊆ R by deﬁnition of J, so IJ

page 18 of Chapter 7

is an integral ideal. Since (using (7.6.8)) I = IR ⊆ IJ ⊆ R, maximality of I implies that either IJ = I or IJ = R. In the latter case, we are ﬁnished, so assume IJ = I. If x ∈ J, then xI ⊆ IJ = I, and by induction, xn I ⊆ I for all n = 1, 2, . . .. Let r be any nonzero element of I. Then rxn ∈ xn I ⊆ I ⊆ R, so R[x] is a fractional ideal. Since R is Noetherian, part (ii) of (7.6.6) implies that R[x] is a ﬁnitely generated R-submodule of K. By (7.1.2), x is integral over R. But R, a Dedekind domain, is integrally closed, so x ∈ R. Therefore J ⊆ R, contradicting (7.6.8). ♣ Problems For Section 7.6 1. Show that a proper ideal P is prime if and only if for all ideals A and B, P ⊇ AB implies that P ⊇ A or P ⊇ B. We are going to show that if an ideal I is contained in the union of the prime ideals P1 , . . . , Pn , then I is contained in some Pi . Equivalently, if for all i = 1, . . . , n, we have I ⊆ Pi , then I ⊆ ∪ni=1 Pi . There is no problem when n = 1, so assume the result holds for n − 1 prime ideals. By the induction hypothesis, for each i there exists xi ∈ I with xi ∈ / Pj , j = i. 2. Show that we can assume withoutloss of generality that xi ∈ Pi for all i. n 3. Continuing Problem 2, let x = i=1 x1 · · · xi−1 xi+1 · · · xn . Show that x ∈ I but x ∈ / n ∪i=1 Pi , completing the proof. 4. If I and J are relatively prime ideals (I + J = R), show that IJ = I ∩ J. More generally, if I1 , . . . , In are relatively prime in pairs (see (2.3.7)), show that I1 · · · In = ∩ni=1 Ii . 5. Show that if a Dedekind domain R is a UFD, then R is a PID. 6. Suppose that in (7.6.9), we would like to invert every maximal ideal of R, rather than the nonzero prime ideals. What is a reasonable hypothesis to add about R? 7. Let R be an integral domain with quotient ﬁeld K. If K is a fractional ideal of R, show that R = K. 8. Let P1 and P2 be relatively prime ideals in the ring R. Show that P1r and P2s are relatively prime for arbitrary positive integers r and s. 7.7 Unique Factorization of Ideals in a Dedekind Domain In the previous section, we inverted nonzero prime ideals in a Dedekind domain. We must now extend this result to nonzero fractional ideals. 7.7.1 Theorem If I is a nonzero fractional ideal of the Dedekind domain R, then I can be factored uniquely as P1n1 P2n2 · · · Prnr where the ni are integers. Consequently, the nonzero fractional ideals form a group under multiplication. Proof. First consider the existence of such a factorization. Without loss of generality, we can restrict to integral ideals. [Note that if r = 0 and rI ⊆ R, then I = (Rr)−1 (rI).] By convention, we regard R as the product of the empty collection of prime ideals, so let S be the set of all nonzero proper ideals of R that cannot be factored in the given form, with all ni positive integers. [This trick will yield the useful result that the factorization of integral ideals only involves positive exponents.] Since R is Noetherian, S, if nonempty, has a maximal element I0 , which is contained in a maximal ideal I. By (7.6.9), I has an inverse fractional ideal J. Thus by (7.6.8) and (7.6.9), I0 = I0 R ⊆ I0 J ⊆ IJ = R. Therefore I0 J is an integral ideal, and we claim that I0 ⊂ I0 J. For if I0 = I0 J, the last paragraph of the proof of (7.6.9) can be reproduced with I replaced by I0 to reach a contradiction. By maximality of I0 , I0 J is a product of prime ideals, say I0 J = P1 · · · Pr (with repetition allowed). Multiply both sides by the prime ideal I to conclude that I0 is a

page 19 of Chapter 7

product of prime ideals, contradicting I0 ∈ S. Thus S must be empty, and the existence of the desired factorization is established. To prove uniqueness, suppose that we have two prime factorizations P1n1 · · · Prnr = Qt11 · · · Qtss where again we may assume without loss of generality that all exponents are positive. [If P −n appears, multiply both sides by P n .] Now P1 contains the product of the Pini , so by (7.6.2), P1 contains Qj for some j. By maximality of Qj , P1 = Qj , and we may renumber so that P1 = Q1 . Multiply by the inverse of P1 (a fractional ideal, but there is no problem) to cancel P1 and Q1 , and continue inductively to complete the proof. ♣ 7.7.2 Corollary A nonzero fractional ideal I is an integral ideal if and only if all exponents in the prime factorization of I are nonnegative. Proof. The “only if” part was noted in the proof of (7.7.1). The “if” part follows because a power of an integral ideal is still an integral ideal. ♣ 7.7.3 Corollary Denote by nP (I) the exponent of the prime ideal P in the factorization of I. (If P does not appear, take nP (I) = 0.) If I1 and I2 are nonzero fractional ideals, then I1 ⊇ I2 if and only if for every prime ideal P of R, nP (I1 ) ≤ nP (I2 ). Proof. We have I2 ⊆ I1 iﬀ I2 I1−1 ⊆ R, and by (7.7.2), this happens iﬀ for every P , nP (I2 ) − nP (I1 ) ≥ 0. 7.7.4 Deﬁnition Let I1 and I2 be nonzero integral ideals. We say that I1 divides I2 if I2 = JI1 for some integral ideal J. Just as with integers, an equivalent statement is that each prime factor of I1 is a factor of I2 . 7.7.5 Corollary If I1 and I2 are nonzero integral ideals, then I1 divides I2 if and only if I1 ⊇ I2 . In other words, for these ideals, DIVIDES MEANS CONTAINS. Proof. By (7.7.4), I1 divides I2 iﬀ nP (I1 ) ≤ nP (I2 ) for every prime ideal P . By (7.7.3), this is equivalent to I1 ⊇ I2 . ♣ The next result explains why Dedekind domains are important in algebraic number theory. 7.7.6 Theorem In the basic AKLB setup of (7.3.9), if A is a Dedekind domain, then so is B. In particular, the ring of algebraic integers in a number ﬁeld is a Dedekind domain. In addition, B is a ﬁnitely generated A-module and the quotient ﬁeld of B is L. Proof. By (7.1.6), B is integrally closed in L. The proof of (7.4.7), with xi replaced by an arbitrary element of L, shows that L is the quotient ﬁeld of B. Therefore B is integrally closed. By (7.5.13), B is a Noetherian ring, and the proof of (7.5.13) shows that B is a Noetherian, hence ﬁnitely generated, A-module. It remains to prove that every nonzero prime ideal Q of B is maximal. Choose any nonzero element x of Q. Since x ∈ B, x satisﬁes a polynomial equation xn + an−1 xn−1 + · · · + a1 x + a0 = 0 with the ai ∈ A. If we take the positive integer n as small as possible, then a0 = 0 by minimality of n. Solving for a0 , we see that a0 ∈ Bx ∩ A ⊆ Q ∩ A, so P = Q ∩ A = 0. But P is the preimage of the prime ideal Q under the inclusion map of A into B. Therefore P is a nonzero prime, hence maximal, ideal of the Dedekind domain A. Consequently, A/P is a ﬁeld.

page 20 of Chapter 7

Now A/P can be identiﬁed with a subring of the integral domain B/Q via y+P → y+Q. Moreover, B/Q is integral over A/P . [B is integral over A, and we can simply use the same equation of integral dependence.] It follows from Section 7.1, Problem 5, that B/Q is a ﬁeld, so Q is a maximal ideal. ♣ Problems For Section 7.7

√ √ √ By (7.2.3), the ring B of algebraic integers in Q( −5) is Z[ −5]. We will show that Z[ −5] is not a unique factorization domain. (For a diﬀerent approach, see Section 2.7, Problems 5-7.) Consider the factorization (1 +

√

−5)(1 −

√

−5) = (2)(3).

1. By computing norms, verify that all four of the above factors are irreducible. 2. Show that the only units of B are ±1. 3. Show that no factor on one side of the above equation is an associate of a factor on the other side, so unique factorization fails. 4. We can use the prime factorization of ideals in a Dedekind domain to compute the greatest common divisor and the least common multiple of two nonzero ideals I and J, exactly as with integers. Show that the greatest common divisor of I and J is I + J and the least common multiple is I ∩ J. 5. A Dedekind domain R comes close to being a PID in the following sense. (All ideals are assumed nonzero.) If I is an integral ideal, in fact if I is a fractional ideal, show that there is an integral ideal J such that IJ is a principal√ideal of R. 6. Show that the ring of algebraic integers in Q( −17) is not a unique factorization domain. 7. In Problem 6, the only algebraic integers √ of norm 1 are ±1. Show that this property does not hold for the algebraic integers in Q( −3). 7.8 Some Arithmetic in Dedekind Domains Unique factorization of ideals in a Dedekind domain permits calculations that are analogous to familiar manipulations involving ordinary integers. In this section, we illustrate some of the ideas. Let P1 , . . . , Pn be distinct nonzero prime ideals of the Dedekind domain R, and let J = P1 · · · Pn . Let Qi be the product of the Pj with Pi omitted, that is, Qi = P1 · · · Pi−1 Pi+1 · · · Pn . (If n = 1, we take Qi = R.) If I is any nonzero ideal of R, then by unique factorization, IQi ⊃ IJ. nFor each i = 1, . . . , n, choose an element ai belonging to IQi but not to IJ, and let a = i=1 ai . 7.8.1 Lemma a ∈ I but for each i, a ∈ / IPi . (In particular, a = 0.) Proof. Since each ai belongs to IQi ⊆ I, we have a ∈ I. Now ai cannot belong to IPi , for if so, ai ∈ IPi ∩ IQi , which is the least common multiple of IPi and IQi (see Section 7.7, Problem 4). But by deﬁnition of Qi , the least common multiple is simply IJ, and this contradicts the choice of ai . We break up the sum deﬁning a as follows: a = (a1 + · · · + ai−1 ) + ai + (ai+1 + · · · + an ).

(1)

If j = i, then aj ∈ IQj ⊆ IPi , so the ﬁrst and third terms of (1) belong to IPi . Since ai ∈ / IPi , as found above, we have a ∈ / IPi . ♣ In Section 7.7, Problem 5, we found that any nonzero ideal is a factor of a principal ideal. We can sharpen this result as follows.

page 21 of Chapter 7

7.8.2 Proposition Let I be a nonzero ideal of the Dedekind domain R. Then there is a nonzero ideal I such that II is a principal ideal (a). Moreover, if J is an arbitrary nonzero ideal of R, I can be chosen to be relatively prime to J. Proof. Let P1 , . . . , Pn be the distinct prime divisors of J, and choose a as in (7.8.1). Then a ∈ I, so (a) ⊆ I. Since divides means contains (see (7.7.5)), I divides (a), so (a) = II for some nonzero ideal I . If I is divisible by Pi , then I = Pi I0 for some nonzero ideal I0 , and (a) = IPi I0 . Consequently, a ∈ IPi , contradicting (7.8.1). ♣ 7.8.3 Corollary A Dedekind domain with only ﬁnitely many prime ideals is a PID. Proof. Let J be the product of all the nonzero prime ideals. If I is any nonzero ideal, then by (7.8.2) there is a nonzero ideal I such that II is a principal ideal (a), with I relatively prime to J. But then the set of prime factors of I is empty, so that I = R. Thus (a) = II = IR = I. ♣ The next result shows that a Dedekind domain is not too far away from a principal ideal domain. 7.8.4 Corollary Let I be a nonzero ideal of the Dedekind domain R, and let a be any nonzero element of I. Then I can be generated by two elements, one of which is a. Proof. Since a ∈ I, we have (a) ⊆ I, so I divides (a), say (a) = IJ. By (7.8.2) there is a nonzero ideal I such that II is a principal ideal (b) and I is relatively prime to J. If gcd stands for greatest common divisor, then the ideal generated by a and b is gcd((a), (b)) = gcd(IJ, II ) = I since gcd(J, I ) = (1). ♣ Problems For Section 7.8 1. Let I(R) be the group of nonzero fractional ideals of a Dedekind domain R. If P (R) is the subset of I(R) consisting of all nonzero principal fractional ideals Rx, x ∈ K, show that P (R) is a subgroup of I(R). The quotient group C(R) = I(R)/P (R) is called the ideal class group of R. Since R is commutative, C(R) is abelian, and it can be shown that C(R) is ﬁnite. 2. Continuing Problem 1, show that C(R) is trivial iﬀ R is a PID. We will now go through the factorization of an ideal in a number ﬁeld. The necessary background is developed in a course in algebraic number theory, but some of the manipulations are √ accessible √ to us now. By (7.2.3), the ring B of algebraic integers of the number ﬁeld Q( −5) is Z[ −5]. (Note that −5 ≡ 3 mod 4.) If we wish to factor the ideal (2) = 2B 2 in B, the the result is x2 +5 ≡ (x+1)2 mod 2. Identifying √ idea is to factor x +5 mod 2, and √ x with −5, we form the ideal P2 = (2, 1 + −5), which turns out to be prime. The desired factorization is (2) = P22 . This technique works if B = Z[α], where the number ﬁeld L is Q(α). √ 3. Show that 1 − −5 ∈ P2 , and conclude that 6 ∈ P22 . 2 . 4. Show that 2 ∈ P22 , hence √ (2) ⊆ P2 √ 2 5. Expand P2 = (2, 1 + −5)(2, 1 + −5), and conclude that P22 ⊆ (2). 6. Following the technique suggested in the above problems, factor x√2 + 5 mod 3, and conjecture that the prime factorization of (3) in the ring of integers of Q( −5) is (3) = P3 P3 for appropriate P3 and P3 . 7. With P3 and P3 as found in Problem 6, verify that (3) = P3 P3 . 7.9 p-adic Numbers We will give a very informal introduction to this basic area of number theory. ( For further details, see Gouvea, “p-adic Numbers”.) Throughout the discussion, p is a ﬁxed prime.

page 22 of Chapter 7

7.9.1 Deﬁnitions and Comments A p-adic integer can be described in several ways. One representation is via a series x = a0 + a1 p + a2 p2 + · · · , ai ∈ Z.

(1)

(Let’s ignore the problem of convergence for now.) The partial sums are xn = a0 + a1 p + · · · + an pn , so that xn − xn−1 = an pn . A p-adic integer can also be deﬁned as a sequence of integers x = {x0 , x1 , . . . , } satisfying xn ≡ xn−1 mod pn , n = 1, 2, . . . .

(2)

Given a sequence satisfying (2), we can recover the coeﬃcients of the series (1) by a0 = x0 , a1 =

x1 − x0 x2 − x1 ,.... , a2 = p p2

The sequences x and y are regarded as deﬁning the same p-adic integer iﬀ xn ≡ yn mod pn+1 , n = 0, 1, . . .. Replacing each xn by the smallest nonnegative integer congruent to it mod pn+1 is equivalent to restricting the ai in (1) to {0, 1, . . . , p − 1}. [We call this the standard representation.] Thus (1) is the limiting case in some sense of an expansion in base p. Sums and products of p-adic integers can be deﬁned by polynomial multiplication if (1) is used. With the representation (2), we take x + y = {xn + yn }, xy = {xn yn }. With addition and multiplication deﬁned in this way, we get the ring of p-adic integers, denoted by θp . (A more common notation is Zp , with the ring of integers modulo p written as Z/pZ. Since the integers mod p occur much more frequently in this text than the p-adic integers, and Zp is a bit simpler than Z/pZ, I elected to use Zp for the integers mod p.) The rational integers Z form a subring of θp via x = {x, x, x, . . .}. We now identify the units of θp . 7.9.2 Proposition The p-adic integer x = {xn } is a unit of θp (also called a p-adic unit) if and only if x0 ≡ 0 mod p. In particular, a rational integer a is a p-adic unit if and only if a ≡ 0 mod p. Proof. If (a0 +a1 p+· · ·)(b0 +b1 p+· · ·) = 1, then a0 b0 = 1, so a0 ≡ 0 mod p, proving the “only if” part. Thus assume that x0 ≡ 0 mod p. By (2), xn ≡ xn−1 ≡ · · · ≡ x0 mod p, so xn ≡ 0 mod p. Therefore xn and pn+1 are relatively prime, so there exists yn such that xn yn ≡ 1 mod pn+1 , hence mod pn . Now by (2), xn ≡ xn−1 mod pn , so xn yn−1 ≡ xn−1 yn−1 ≡ 1 mod pn . Thus xn yn ≡ xn yn−1 mod pn , so yn ≡ yn−1 mod pn . The sequence y = {yn } is therefore a p-adic integer, and by construction, xy = 1. ♣ 7.9.3 Corollary Every nonzero p-adic integer has the form x = pn u where n ≥ 0 and u is a p-adic unit. Consequently, θp is an integral domain. Furthermore, θp has only one prime element p, and every x ∈ θp is a power of p, up to multiplication by a unit. Proof. The series representation for x has a nonzero term an pn of lowest degree n, where an can be taken between 1 and p − 1. Factor out pn to obtain x = pn u, where u is a unit by (7.9.2). ♣ 7.9.4 Deﬁnitions and Comments The quotient ﬁeld Qp of θp is called the ﬁeld of padic numbers. By (7.9.3), each α ∈ Qp has the form pm u, where m is an integer (possibly negative) and u is a unit in θp . Thus α has a “Laurent expansion” a−r a−1 + ··· + + a0 + a1 p + · · · . pr p

page 23 of Chapter 7

Another representation is α = x/pr , where x is a p-adic integer and r ≥ 0. This version is convenient for doing addition and multiplication in Qp . The rationals Q are a subﬁeld of Qp . To see this, let a/b be a rational number in lowest terms (a and b relatively prime). If p does not divide b, then by (7.9.2), b is a unit of θp . Since a ∈ Z ⊆ θp , we have a/b ∈ θp . If b = pt b where p does not divide b , we can factor out pt and reduce to the previous case. Thus a/b always belongs to Qp , and a/b ∈ θp iﬀ p does not divide b. Rational numbers belonging to θp are sometimes called p-integers. We now outline a procedure for constructing the p-adic numbers formally. 7.9.5 Deﬁnitions and Comments The p-adic valuation on Qp is deﬁned by vp (pm u) = m. In general, a valuation v on a ﬁeld F is a real-valued function on F \ {0} satisfying: (a) v(xy) = v(x) + v(y); (b) v(x + y) ≥ min(v(x), v(y)). By convention, we take v(0) = + ∞. The representation x = pm u shows that vp is indeed a valuation on Qp . If c is any real number greater than 1, then the valuation v induces an absolute value on F , namely, |x| = c−v(x) . When v = vp , the constant c is usually taken to be p, and we obtain the p-adic absolute value |x|p = p−vp (x) . Thus the p-adic absolute value of pn is p−n , which approaches 0 exponentially as n approaches inﬁnity. In general, an absolute value on a ﬁeld F is a real-valued function on F such that: (i) |x| ≥ 0, with equality if and only if x = 0; (ii) |xy| = |x||y|; (iii) |x + y| ≤ |x| + |y|. By (b), an absolute value induced by a valuation satisﬁes a property that is stronger than (iii): (iv) |x + y| ≤ max(|x|, |y|). An absolute value satisfying (iv) is said to be nonarchimedian. 7.9.6 Proposition Let F be the quotient ﬁeld of an integral domain R. The absolute value | | on F is nonarchimedian if and only if |n| ≤ 1 for every integer n = 1 ± · · · ± 1 ∈ R. Proof. Assume a nonarchimedian absolute value. By property (ii) of (7.9.5), | ± 1| = 1. If |n| ≤ 1, then by property (iv), |n ± 1| ≤ 1, and the desired conclusion follows by induction. Conversely, assume that the absolute value of every integer is at most 1. To prove (iv), it suﬃces to show that |x + 1| ≤ max(|x|, 1) for every x ∈ F . [If y = 0 in (iv), divide by |y|.] By the binomial theorem, |x + 1| = | n

n n r=0

n n x |≤ | ||x|r . r r r=0 r

page 24 of Chapter 7

By hypothesis, the integer nr has absolute value at most 1. If |x| > 1, then |x|r ≤ |x|n for all r = 0, 1, . . . , n. If |x| ≤ 1, then |x|r ≤ 1. Consequently, |x + 1|n ≤ (n + 1)max(|x|n , 1). Take nth roots and let n → ∞ to get |x + 1| ≤ max(|x|, 1). ♣ The next result may seem innocuous, but it leads to a remarkable property of nonarchimedian absolute values. 7.9.7 Proposition If | | is a nonarchimedian absolute value, then |x| = |y| implies |x + y| = max(|x|, |y|). Proof. First note that | − y| = |(−1)y| = | − 1||y| = |y|. We can assume without loss of generality that |x| > |y|. Using property (iv) of (7.9.5), we have |x| = |x + y − y| ≤ max(|x + y|, |y|) = |x + y|. [Otherwise, max(|x + y|, |y|) = |y|, hence |x| ≤ |y| < |x|, a contradiction.] Since |x + y| ≤ max(|x|, |y|) = |x|, the result follows. ♣ Any absolute value determines a metric via d(x, y) = |x − y|. This distance function can be used to measure the length of the sides of a triangle. 7.9.8 Corollary With respect to the metric induced by a nonarchimedian absolute value, all triangles are isosceles. Proof. Let the vertices of the triangle be x, y and z. Then |x − y| = |(x − z) + (z − y)|. If |x − z| = |z − y|, then two side lengths are equal. If |x − z| = ||z − y|, then by (7.9.7), |x − y| = max(|x − z|, |z − y|), and again two side lengths are equal. ♣ We now look at the p-adic numbers from the viewpoint of valuation theory. 7.9.9 Deﬁnitions and Comments Let | | be a nonarchimedian absolute value on the ﬁeld F . The valuation ring of | | is θ = {x ∈ F : |x| ≤ 1}. In the p-adic case, θ = {x ∈ Qp : vp (x) ≥ 0} = θp . By properties (ii) and (iv) of (7.9.5), θ is a subring of F . The valuation ideal of | | is β = {x ∈ F : |x| < 1}. In the p-adic case, β = {x ∈ Qp : vp (x) ≥ 1} = pθp , those p-adic integers whose series representation has no constant term. To verify that β is an ideal of θ, note that if x, y ∈ β and r ∈ θ, then |rx| = |r||x| ≤ |x| < 1 and |x + y| ≤ max(|x|, |y|) < 1. Now if x ∈ θ \ β, then |x| = 1, hence |x−1 | = 1/|x| = 1, so x−1 ∈ θ and x is a unit of θ. On the other hand, if x ∈ β, then x cannot be a unit of θ. [If xy = 1, then 1 = |x||y| ≤ |x| < 1, a contradiction.] Thus the ideal β is the set of all nonunits of θ. No proper ideal I of θ can contain a unit, so I ⊆ β. It follows that β is the unique maximal ideal of θ. A ring with a unique maximal ideal is called a local ring. We will meet such rings again when we examine the localization process in Section 8.5.

page 25 of Chapter 7

To construct the p-adic numbers, we start with the p-adic valuation on the integers, and extend it to the rationals in the natural way: vp (a/b) = vp (a) − vp (b). The p-adic valuation then determines the p-adic absolute value, which induces a metric d on Q. [Because d comes from a nonarchimedian absolute value, it will satisfy the ultrametric inequality d(x, y) ≤ max(d(x, z), d(z, y)), which is stronger than the triangle inequality.] The process of constructing the real numbers by completing the rationals using equivalence classes of Cauchy sequences is familiar. The same process can be carried out using the p-adic absolute value rather than the usual absolute value on Q. The result is a complete metric space, the ﬁeld of p-adic numbers, in which Q is dense. Ostrowski’s theorem says that the usual absolute value | |∞ on Q, along with the p-adic absolute values | |p for all primes p, and the trivial absolute value (|0| = 0; |x| = 1 for x = 0), essentially exhaust all possibilities. To be more precise, two absolute values on a ﬁeld F are equivalent if the corresponding metrics on F induce the same topology. Any nontrivial absolute value on Q is equivalent to | |∞ or to one of the | |p . Problems For Section 7.9 1. Take p = 3, and compute the standard representation of (2 + p + p2 )(2 + p2 ) in two ways, using (1) and (2) of (7.9.1). Check the result by computing the product using ordinary multiplication of two integers, and then expanding in base p = 3. 2. Express the p-adic integer -1 as an inﬁnite series of the form (1), using the standard representation. 3. Show that every absolute value on a ﬁnite ﬁeld is trivial. 4. Show that an absolute value is archimedian iﬀ the set S = {|n| : n ∈ Z} is unbounded. 5. Show that a ﬁeld that has anarchimedian absolute value must have characteristic 0. 6. Show that an inﬁnite series zn of p-adic numbers converges if and only if zn → 0 as n → ∞. 7. Show that the sequence an = n! of p-adic integers converges to 0. 8. Does the sequence an = n converge in Qp ?

page 1 of Chapter 8

CHAPTER 8 INTRODUCING ALGEBRAIC GEOMETRY (COMMUTATIVE ALGEBRA 2) We will develop enough geometry to allow an appreciation of the Hilbert Nullstellensatz, and look at some techniques of commutative algebra that have geometric signiﬁcance. As in Chapter 7, unless otherwise speciﬁed, all rings will be assumed commutative. 8.1 Varieties 8.1.1 Deﬁnitions and Comments We will be working in k[X1 , . . . , Xn ], the ring of polynomials in n variables over the ﬁeld k. (Any application of the Nullstellensatz requires that k be algebraically closed, but we will not make this assumption until it becomes necessary.) The set An = An (k) of all n-tuples with components in k is called aﬃne n-space. If S is a set of polynomials in k[X1 , . . . , Xn ], then the zero-set of S, that is, the set V = V (S) of all x ∈ An such that f (x) = 0 for every f ∈ S, is called a variety. (The term “aﬃne variety” is more precise, but we will use the short form because we will not be discussing projective varieties.) Thus a variety is the solution set of simultaneous polynomial equations. If I is the ideal generated by S, then I consists of all ﬁnite linear combinations gi fi with gi ∈ k[X1 , . . . , Xn ] and fi ∈ S. It follows that V (S) = V (I), so every variety is the variety of some ideal. We now prove that we can make An into a topological space by taking varieties as the closed sets. 8.1.2 Proposition

(1) If Vα = V (Iα ) for all α ∈ T , then Vα = V ( Iα ). Thus an arbitrary intersection of varieties is a variety. r (2) If Vj = V (Ij ), j = 1, . . . , r, then j=1 Vj = V ({f1 · · · fr : fj ∈ Ij , 1 ≤ j ≤ r}). Thus a ﬁnite union of varieties is a variety. (3) An = V (0) and ∅ = V (1), so the entire space and the empty set are varieties. Consequently, there is a topology on An , called the Zariski topology, such that the closed sets and the varieties coincide. Proof.

(1) If x ∈ An , then x ∈ Vα iﬀ every polynomial in every Iα vanishes at x iﬀ x ∈ V ( Iα ). r (2) x ∈ j=1 Vj iﬀ for some j, every fj ∈ Ij vanishes at x iﬀ x ∈ V ({f1 · · · fr : fj ∈ Ij for all j}). (3) The zero polynomial vanishes everywhere and a nonzero constant polynomial vanishes nowhere. ♣ Note that condition (2) can also be expressed as ∪rj=1 Vj = V (

r

Ij ) = V (∩rj=1 Ij ).

j=1

[See (7.6.1) for the deﬁnition of a product of ideals.] We have seen that every subset of k[X1 , . . . , Xn ], in particular every ideal, determines a variety. We can reverse this process as follows. 8.1.3 Deﬁnitions and Comments If X is an arbitrary subset of An , we deﬁne the ideal of X as I(X) = {f ∈ k[X1 , . . . , Xn ] : f vanishes on X}. By deﬁnition we have: (4) If X ⊆ Y then I(X) ⊇ I(Y );

if S ⊆ T then V (S) ⊇ V (T ).

Now if S is any set of polynomials, deﬁne IV (S) as I(V (S)), the ideal of the zero-set of S; we are simply omitting parentheses for convenience. Similarly, if X is any subset of An , we can deﬁne V I(X), IV I(X), V IV (S), and so on. From the deﬁnitions we have:

page 2 of Chapter 8

(5) IV (S) ⊇ S; V I(X) ⊇ X. [If f ∈ S then f vanishes on V (S), hence f ∈ IV (S). If x ∈ X then every polynomial in I(X) vanishes at x, so x belongs to the zero-set of I(X).] If we keep applying V ’s and I’s alternately, the sequence stabilizes very quickly: (6) V IV (S) = V (S); IV I(X) = I(X). [In each case, apply (4) and (5) to show that the left side is a subset of the right side. If x ∈ V (S) and f ∈ IV (S) then f (x) = 0, so x ∈ V IV (S). If f ∈ I(X) and x ∈ V I(X) then x belongs to the zero-set of I(X), so f (x) = 0. Thus f vanishes on V I(X), so f ∈ IV I(X).] Since every polynomial vanishes on the empty set (vacuously), we have: (7) I(∅) = k[X1 , . . . , Xn ]. The next two properties require a bit more eﬀort. (8) If k is an inﬁnite ﬁeld, then I(An ) = {0}; (9) If x = (a1 , . . . , an ) ∈ An , then I({x}) = (X1 − a1 , . . . , Xn − an ). Property (8) holds for n = 1 since a nonconstant polynomial in one variable has only ﬁnitely many zeros. Thus f = 0 implies that f ∈ / I(A1 ). If n > 1, let f = ar X1r + · · · + a1 X1 + a0 where the ai are polynomials in X2 , . . . , Xn and ar = 0. By the induction hypothesis, there is a point (x2 , . . . , xn ) at which ar does not vanish. Fixing this point, we can regard f as a polynomial in X1 , which cannot possibly vanish at all x1 ∈ k. Thus f ∈ / I(An ). To prove (9), note that the right side is contained in the left side because Xi − ai is 0 when Xi = ai . Also, the result holds for n = 1 by the remainder theorem (2.5.2). Thus assume n > 1 and let f = br X1r + · · · + b1 X1 + b0 ∈ I({x}), where the bi are polynomials in X2 , . . . , Xn and br = 0. By the division algorithm (2.5.1), we have f = (X1 − a1 )g(X1 , . . . , Xn ) + h(X2 , . . . , Xn ) and h must vanish at (a2 , . . . , an ). By the induction hypothesis, h ∈ (X2 − a2 , . . . , Xn − an ), hence f ∈ (X1 − a1 , X2 − a2 , . . . , Xn − an ). Problems For Section 8.1 A variety is said to be reducible if it can be expressed as the union of two proper subvarieties; otherwise the variety is irreducible. In Problems 1-4, we are going to show that a variety V is irreducible if and only if I(V ) is a prime ideal. 1. Assume that I(V ) is not prime, and let f1 f2 ∈ I(V ) with f1 , f2 ∈ / I(V ). If x ∈ V , show that x ∈ / V (f1 ) implies x ∈ V (f2 ) (and similarly, x ∈ / V (f2 ) implies x ∈ V (f1 )). 2. Show that V is reducible. 3. Show that if V and W arevarieties with V ⊂ W , then I(V ) ⊃ I(W ). 4. Now assume that V = V1 V2 , with V1 , V2 ⊂ V . By Problem 3, we can choose fi ∈ I(Vi ) with fi ∈ / I(V ). Show that f1 f2 ∈ I(V ), so I(V ) is not a prime ideal. 5. Show that any variety is the union of ﬁnitely many irreducible subvarieties. 6. Show that the decomposition of Problem 5 is unique, assuming that we discard any subvariety that is contained in another one. 7. Assume that k is algebraically closed. Suppose that An is covered by open sets An \ V (Ii ) Ii , the set of in the Zariski topology. Let I is the ideal generated by the Ii , so that I = all ﬁnite sums xi1 + · · · xir with xij ∈ Iij . Show that 1 ∈ I. (You may appeal to the weak Nullstellensatz, to be proved in Section 8.4.) 8. Show that An is compact in the Zariski topology.

page 3 of Chapter 8

8.2 The Hilbert Basis Theorem If S is a set of polynomials in k[X1 , . . . , Xn ], we have deﬁned the variety V (S) as the zero-set of S, and we know that V (S) = V (I), where I is the ideal generated by S. Thus any set of simultaneous polynomial equations deﬁnes a variety. In general, inﬁnitely many equations may be involved, but as Hilbert proved, an inﬁnite collection of equations can always be replaced by a ﬁnite collection. The reason is that every ideal of k[X1 , . . . , Xn ] has a ﬁnite set of generators, in other words, k[X1 , . . . , Xn ] is a Noetherian ring. The ﬁeld k is, in particular, a PID, so k is Noetherian. The key step is to show that if R is a Noetherian ring, then the polynomial ring in n variables over R is also Noetherian. 8.2.1 Hilbert Basis Theorem If R is a Noetherian ring, then R[X1 , . . . , Xn ] is also Noetherian. Proof. By induction, we can assume n = 1. Let I be an ideal of R[X], and let J be the ideal of all leading coeﬃcients of polynomials in I. (The leading coeﬃcient of 5X 2 −3X +17 is 5; the leading coeﬃcient of the zero polynomial is 0.) By hypothesis, J is ﬁnitely generated, say by a1 , . . . , an . Let fi be a polynomial in I whose leading coeﬃcient is ai , and let di be the degree of fi . Let I ∗ consist of all polynomials in I of degree at most d = max {di : 1 ≤ i ≤ n}. Then I ∗ is an R-submodule of the free R-module M of all polynomials b0 +b1 X +· · ·+bd X d , bi ∈ R. Now a ﬁnitely generated free R-module is a ﬁnite direct sum of copies of R, hence M , and therefore I ∗ , is Noetherian. Thus I ∗ can be generated by ﬁnitely many polynomials g1 , . . . , gm . Take I0 to be the ideal of R[X] generated by f1 , . . . , fn , g1 , . . . , gm . We will show that I0 = I, proving that I is ﬁnitely generated. First observe that fi ∈ I and gj ∈ I ∗ ⊆ I, so I0 ⊆ I. Thus we must show that each h ∈ I belongs to I0 . Case 1: deg h ≤ d Then h ∈ I ∗ , so h is a linear combination of the gj (with coeﬃcients in R ⊆ R[X]), so h ∈ I0 . Case 2: deg h = r > d Let a be the leading coeﬃcient of h. Since a ∈ J, we have a = ci ∈ R. Let n q =h− ci X r−di fi ∈ I.

n

i=1 ci ai

with the

i=1 r

The coeﬃcient of X in q is a−

n

ci ai = 0

i=1

so that deg q < r. We can iterate this degree-reduction process until the resulting polynomial has degree d or less, and therefore belongs to I0 . But then h is a ﬁnite linear combination of the fi and gj . ♣ 8.2.2 Corollary Every variety is the intersection of ﬁnitely many hypersurfaces (zero-sets of single polynomials). Proof. Let V = V (I) be a variety. By (8.2.1), I has ﬁnitely many generators f1 , . . . , fr . r But then V = i=1 V (fi ). ♣ 8.2.3 Formal Power Series The argument used to prove the Hilbert basis theorem can be adapted to show that if R is Noetherian, then the ring R[[X]] of formal power series is Noetherian. We cannot simply reproduce the proof because an inﬁnite series has no term of highest degree, but we can look at the lowest degree term. If f = ar X r + ar+1 X r+1 + · · ·, where r is a nonnegative

page 4 of Chapter 8

integer and ar = 0, let us say that f has degree r and leading coeﬃcient ar . (If f = 0, take the degree to be inﬁnite and the leading coeﬃcient to be 0.) If I is an ideal of R[[X]], we must show that I is ﬁnitely generated. We will inductively construct a sequence of elements fi ∈ R[[X]] as follows. Let f1 have minimal degree among elements of I. Suppose that we have chosen f1 , . . . , fi , where fi has degree di and leading coeﬃcient ai . We then select fi+1 satisfying the following three requirements: 1. fi+1 belongs to I; 2. ai+1 does not belong to (a1 , . . . , ai ), the ideal of R generated by the aj , j = 1, . . . , i; 3. Among all elements satisfying the ﬁrst two conditions, fi+1 has minimal degree. The second condition forces the procedure to terminate in a ﬁnite number of steps; otherwise there would be an inﬁnite ascending chain (a1 ) ⊂ (a1 , a2 ) ⊂ (a1 , a2 , a3 ) ⊂ · · ·. If stabilization occurs at step k, we will show that I is generated by f1 , . . . , fk . Let g = aX d + · · · be an element of I of degree d and leading coeﬃcient a. Then a ∈ (a1 , . . . , ak ) (Problem 1). Case 1: d ≥ dk . Since di ≤ di+1 for all i (Problem 2), we have d ≥ di for i = 1, . . . , k. Now k a = i=1 ci0 ai with the ci0 ∈ R. Deﬁne g0 =

k

ci0 X d−di fi

i=1

so that g0 has degree d and leading coeﬃcient a, and consequently rg − g0 has degree greater than d. Having deﬁned g0 , . . . , gr ∈ (f1 , . . . , fk ) such that g − i=0 gi has degree greater than d + r, say r g− gi = bX d+r+1 + . . . . i=0

(The argument is the same if the degree is greater than d + r + 1.) Now b ∈ (a1 , . . . , ak ) (Problem 1 again), so k b= ci,r+1 ai i=1

with ci,r+1 ∈ R. We deﬁne gr+1 =

k

ci,r+1 X d+r+1−di fi

i=1

so that g −

r+1 i=0

gi has degree greater than d + r + 1. Thus g=

∞ r=0

gr =

∞ k

cir X d+r−di fi

r=0 i=1

and it follows upon reversing the order of summation that g ∈ (f1 , . . . , fk ). (The reversal is legal because the inner summation is ﬁnite. For a given nonnegative integer j, there are only ﬁnitely many terms of the form bX j .) Case 2: d < dk . As above, a ∈ (a1 , . . . , ak ), so there is a smallest m between 1 and k such that a ∈ (a1 , . . . , am ). It follows that d ≥ dm (Problem 3). As in case 1 we have m a = i=1 ci ai with ci ∈ R. Deﬁne h=

m i=1

ci X d−di fi ∈ (f1 , . . . , fk ) ⊆ I.

page 5 of Chapter 8

The leading coeﬃcient of h is a, so the degree of g − h is greater than d. We replace g by g−h and repeat the procedure. After at most dk − d iterations, we produce an element g − hi in I of degree at least dk , with all hi ∈ (f1 , . . . , fk ). By the analysis of case 1, g ∈ (f1 , . . . , fk ). Problems For Section 8.2 1. Justify the step a ∈ (a1 , . . . , ak ) in (8.2.3). 2. Justify the step di ≤ di+1 in (8.2.3). 3. Justify the step d ≥ dm in (8.2.3). 4. Let R be a subring of the ring S, and assume that S is ﬁnitely generated as an algebra over R. In other words, there are ﬁnitely many elements x1 , . . . , xn ∈ S such that the smallest subring of S containing the xi and all elements of R is S itself. Show that S is a homomorhic image of the polynomial ring R[X1 , . . . , Xn ]. 5. Continuing Problem 4, show that if R is Noetherian, then S is also Noetherian. 8.3 The Nullstellensatz: Preliminaries We have observed that every variety V deﬁnes an ideal I(V ) and every ideal I deﬁnes a variety V (I). Moreover, if I(V1 ) = I(V2 ), then V1 = V2 by (6) of (8.1.3). But it is entirely possible for many ideals to deﬁne the same variety. For example, the ideals (f ) and (f m ) need not coincide, but their zero-sets are identical. Appearances to the contrary, the two statements in part (6) of (8.1.3) are not symmetrical. A variety V is, by deﬁnition, always expressible as V (S) for some collection S of polynomials, but an ideal I need not be of the special form I(X). Hilbert’s Nullstellensatz says that if two ideals deﬁne the same variety, then, informally, the ideals are the same “up to powers”. More precisely, if g belongs to one of the ideals, then g r belongs to the other ideal for some positive integer r. Thus the only factor preventing a one-to-one correspondence between ideals and varieties is that a polynomial can be raised to a power without aﬀecting its zero-set. In this section we collect some results needed for the proof of the Nullstellensatz. We begin by showing that each point of An determines a maximal ideal. 8.3.1 Lemma If a = (a1 , . . . , an ) ∈ An , then I = (X1 − a1 , . . . , Xn − an ) is a maximal ideal. Proof. Suppose that I is properly contained in the ideal J, with f ∈ J \ I. Apply the division algorithm n times to get f = A1 (X1 − a1 ) + A2 (X2 − a2 ) + . . . + An (Xn − an ) + b where A1 ∈ k[X1 , . . . , Xn ], A2 ∈ k[X2 , . . . , Xn ], . . . , An ∈ k[Xn ], b ∈ k. Note that b cannot be 0 since f ∈ / I. But f ∈ J, so by solving the above equation for b we have b ∈ J, hence 1 = (1/b)b ∈ J. Consequently, J = k[X1 , . . . , Xn ]. ♣ The following deﬁnition will allow a precise statement of the Nullstellensatz. 8.3.2 Deﬁnition The radical of an ideal I (in any commutative ring R) is the set of all elements f ∈ R such that f r ∈ I for some positive integer r. √ r A popular notation for the radical of I is I. If f√ and g s belong to I, then by the r+s−1 binomial theorem, (f + g) ∈ I, and it follows that I is an ideal. √ 8.3.3 Lemma If I is any ideal of k[X1 , . . . , Xn ], then I ⊆ IV (I). √ Proof. If f ∈ I, then f r ∈ I for some positive integer r. But then f r vanishes on V (I), hence so does f . Therefore f ∈ IV (I). ♣ √ √ The Nullstellensatz states that IV (I) = I, and the hard part is to prove that IV (I) ⊆ I. The technique is known as the “Rabinowitsch trick”, and it is indeed very clever.

page 6 of Chapter 8

Assume that f ∈ IV (I). We introduce a new variable Y and work in k[X1 , . . . , Xn , Y ]. If I is an ideal of k[X1 , . . . , Xn ], then by the Hilbert basis theorem, I is ﬁnitely generated, say by f1 , . . . , fm . Let I ∗ be the ideal of k[X1 , . . . , Xn , Y ] generated by f1 , . . . , fm , 1 − Y f . [There is a slight ambiguity: by fi (X1 , . . . , Xn , Y ) we mean fi (X1 , . . . , Xn ), and similarly for f .] At an appropriate moment we will essentially set Y equal to 1/f and come back to the original problem. 8.3.4 Lemma If (a1 , . . . , an , an+1 ) is any point in An+1 and (a1 , . . . , an ) ∈ V (I) (in other words, the fi , i = 1, . . . , m, vanish at (a1 , . . . , an )), then (a1 , . . . , an , an+1 ) ∈ / V (I ∗ ). Proof. We are assuming that f ∈ IV (I), so that f vanishes on the zero-set of {f1 , . . . , fm }. In particular, f (a1 , . . . , an ) = 0. The value of 1 − Y f at (a1 , . . . , an , an+1 ) is therefore 1 − an+1 f (a1 , . . . , an ) = 1 − an+1 (0) = 1 = 0. But 1 − Y f ∈ I ∗ , so (a1 , . . . , an , an+1 ) does not belong to the zero-set of I ∗ . ♣ 8.3.5 Lemma If (a1 , . . . , an , an+1 ) is any point in An+1 and (a1 , . . . , an ) ∈ / V (I), then / V (I ∗ ). Consequently, by (8.3.4), V (I ∗ ) = ∅. (a1 , . . . , an , an+1 ) ∈ Proof. By hypothesis, fi (a1 , . . . , an , an+1 ) = 0 for some i, and since fi ∈ I ∗ , (a1 , . . . , an+1 ) cannot belong to the zero-set of I ∗ . ♣ At this point we are going to assume what is called the weak Nullstellensatz, namely that if I is a proper ideal of k[X1 , . . . , Xn ], then V (I) is not empty. 8.3.6 Lemma There are polynomials g1 , . . . , gm , h ∈ k[X1 , . . . , Xn , Y ] such that 1=

m

gi fi + h(1 − Y f ).

(1)

i=1

This equation also holds in the rational function ﬁeld k(X1 , . . . , Xn , Y ) consisting of quotients of polynomials in k[X1 , . . . , Xn , Y ]. Proof. By (8.3.4) and (8.3.5), V (I ∗ ) = ∅, so by the weak Nullstellensatz, I ∗ = k[X1 , . . . , Xn , Y ]. In particular, 1 ∈ I ∗ , and since I ∗ is generated by f1 , . . . , fm , 1 − Y f , there is an equation of the speciﬁed form. The equation holds in the rational function ﬁeld because a polynomial is a rational function. ♣ 8.3.7 The Rabinowitsch Trick The idea is to set Y = 1/f , so that (1) becomes 1=

m

gi (X1 , . . . , Xn , 1/f (X1 , . . . , Xn ))fi (X1 , . . . , Xn ).

(2)

i=1

√ Is this legal? First of all, if f is the zero polynomial, then certainly f ∈ I, so we can assume f = 0. To justify replacing Y by 1/f , consider the ring homomorphism from k[X1 , . . . , Xn , Y ] to k(X1 , . . . , Xn ) determined by Xi → Xi , i = 1, . . . , n, Y → 1/f (X1 , . . . , Xn ). Applying this mapping to (1), we get (2). Now the right side of (2) is a sum of rational functions whose denominators are various powers of f . If f r is the highest power that appears, we can absorb all denominators by multiplying (2) by f r . The result is an equation of the form r

f =

m

hi (X1 , . . . , Xn )fi (X1 , . . . , Xn )

i=1

where the hi are polynomials in k[X1 , . . . , Xn ]. Consequently, f r ∈ I. ♣ The ﬁnal ingredient is a major result in its own right.

page 7 of Chapter 8

8.3.8 Noether Normalization Lemma Let A be a ﬁnitely generated k-algebra, where k is a ﬁeld. In other words, there are ﬁnitely many elements x1 , . . . , xn in A that generate A over k in the sense that every element of A is a polynomial in the xi . Equivalently, A is a homomorphic image of the polynomial ring k[X1 , . . . , Xn ] via the map determined by Xi → xi , i = 1, . . . , n. There exists a subset {y1 , . . . , yr } of A such that the yi are algebraically independent over k and A is integral over k[y1 , . . . , yr ]. Proof. Let {x1 , . . . , xr } be a maximal algebraically independent subset of {x1 , . . . , xn }. If n = r we are ﬁnished, since we can take yi = xi for all i. Thus assume n > r, in which case x1 , . . . , xn are algebraically dependent over k. Thus there is a nonzero polynomial f ∈ k[X1 , . . . , Xn ] such that f (x1 , . . . , xn ) = 0. We can assume n > 1, for if n = 1 and r = 0, then A = k[x1 ] and we can take {y1 , . . . , yr } to be the empty set. We ﬁrst assume that k is inﬁnite and give a proof by induction on n. (It is possible to go directly to the general case, but the argument is not as intricate for an inﬁnite ﬁeld.) Decompose f into its homogeneous components (sums of monomials of the same degree). Say that g is the homogeneous component of maximum degree d. Then, regarding g as a polynomial in Xn whose coeﬃcients are polynomials in the other Xi , we have, relabeling variables if necessary, g(X1 , . . . , Xn−1 , 1) = 0. Since k is inﬁnite, it follows from (8.1.3) part (8) that there are elements a1 , . . . , an−1 ∈ k such that g(a1 , . . . , an−1 , 1) = 0. Set zi = xi − ai xn , i = 1, . . . , n − 1, and plug into f (x1 , . . . , xn ) = 0 to get an equation of the form g(a1 , . . . , an−1 , 1)xdn + terms of degree less than d in xn = 0. A concrete example may clarify the idea. If f (x1 , x2 ) = g(x1 , x2 ) = x21 x32 and x1 = z1 +a1 x2 , then the substitution yields (z12 + 2a1 z1 x2 + a21 x22 )x32 which indeed is g(a1 , 1)x52 plus terms of degree less than 5 in x2 . Divide by g(a1 , . . . , an−1 , 1) = 0 to conclude that xn is integral over B = k[z1 , . . . , zn−1 ]. By the induction hypothesis, there are elements y1 , . . . , yr algebraically independent over k such that B is integral over k[y1 , . . . , yr ]. But the xi , i < n, are integral over B since xi = zi + ai xn . By transitivity (see (7.1.4)), x1 , . . . , xn are integral over k[y1 , . . . , yr ]. Thus (see (7.1.5)) A is integral over k[y1 , . . . , yr ]. Now we consider arbitrary k. As before, we produce a nonzero polynomial f such that f (x1 , . . . , xn ) = 0. We assign a weight wi = sn−i to the variable Xi , where s is a large positive integer. (It suﬃces to take s greater than the total degree of f , that is, the a1 an sum of the degrees of all monomials of f , we n in f .) If h = λX1 · · · Xn is a monomial deﬁne the weight of h as w(h) = i=1 ai wi . The point is that if h = µX1b1 · · · Xnbn , then w(h) > w(h ) iﬀ h > h in the lexicographic ordering, that is, for some m we have ai = bi for i ≤ m, and am+1 > bm+1 . We take h to be the monomial of maximum weight. (If two monomials diﬀer in the lexicographic ordering, they must have diﬀerent weights.) Set i zi = xi − xw n , 1 ≤ i ≤ n − 1, and plug into f (x1 , . . . , xn ) = 0 to get cxnw(h) + terms of lower degree in xn = 0. 1 For example, if f (x1 , x2 ) = h(x1 , x2 ) = x31 x22 , then x1 = z1 + xw 2 gives

2w1 2 1 1 (z13 + 3z12 xw + x3w 2 + 3z1 x2 2 )x2

and w(h) = 3w1 + 2w2 = 3w1 + 2 since sn−2 = s0 = 1. Thus xn is integral over B = k[z1 , . . . , zn−1 ], and an induction argument ﬁnishes the proof as in the ﬁrst case. ♣ 8.3.9 Corollary Let B be a ﬁnitely generated k-algebra, where k is a ﬁeld. If I is a maximal ideal of B, then B/I is a ﬁnite extension of k.

page 8 of Chapter 8

Proof. The ﬁeld k can be embedded in B/I via c → c + I, c ∈ k. [If c ∈ I, c = 0, then c−1 c = 1 ∈ I, a contradiction.] Since A = B/I is also a ﬁnitely generated k-algebra, it follows from (8.3.8) that there is a subset {y1 , . . . , yr } of A with the yi algebraically independent over k and A integral over k[y1 , . . . , yr ]. Now A is a ﬁeld (because I is a maximal ideal), and therefore so is k[y1 , . . . , yr ] (see the Problems in Section 7.1). But this will lead to a contradiction if r ≥ 1, because 1/y1 ∈ / k[y1 , . . . , yr ]. (If 1/y1 = g(y1 , . . . , yr ) ∈ k[y1 , . . . , yr ], then y1 g(y1 , . . . , yr ) = 1, contradicting algebraic independence.) Thus r must be 0, so A is integral, hence algebraic, over the ﬁeld k. Therefore A is generated over k by ﬁnitely many algebraic elements, so by (3.3.3), A is a ﬁnite extension of k. ♣ 8.3.10 Corollary Let A be a ﬁnitely generated k-algebra, where k is a ﬁeld. If A is itself a ﬁeld, then A is a ﬁnite extension of k. Proof. As in (8.3.9), with B/I replaced by A. ♣ Problems For Section 8.3 1. Let S be a multiplicative subset of the ring R (see (2.8.1)). If I is an ideal that is disjoint from S, then by Zorn’s lemma, there is an ideal J that is maximal among ideals disjoint from S. Show that J must be prime. 2. Show that the radical of the ideal I is the intersection of all√prime ideals containing I. [If f r ∈ I ⊆ P , P prime, then f ∈ P . Conversely, assume f ∈ / I. With a clever choice of multiplicative set S, show that for some prime ideal P containing I, we have f ∈ / P .] 3. An algebraic curve is a variety deﬁned by a nonconstant polynomial in two variables. Show (using the Nullstellensatz) that the polynomials f and g deﬁne the same algebraic curve iﬀ f divides some power of g and g divides some power of f . Equivalently, f and g have the same irreducible factors. 4. Show that the variety V deﬁned over the complex numbers by the two polynomials Y 2 − XZ and Z 2 − X 2 Y is the union of the line L given by Y = Z = 0, X arbitrary, and the set W of all (t3 , t4 , t5 ), t ∈ C. 5. The twisted cubic is the variety V deﬁned over the complex numbers by Y − X 2 and Z − X 3 . In parametric form, V = {(t, t2 , t3 ) : t ∈ C}. Show that V is irreducible. [The same argument works for any variety that can be parametrized over an inﬁnite ﬁeld.] 6. Find parametrizations of the following algebraic curves over the complex numbers. (It is permissible for your parametrizations to fail to cover ﬁnitely many points of the curve.) (a) The unit circle x2 + y 2 = 1; (b) The cuspidal cubic y 2 = x3 ; (c) The nodal cubic y 2 = x2 + x3 . 7. Let f be an irreducible polynomial, and g an arbitrary polynomial, in k[x, y]. If f does not divide g, show that the system of simultaneous equations f (x, y) = g(x, y) = 0 has only ﬁnitely many solutions. 8.4 The Nullstellensatz: Equivalent Versions And Proof We are now in position to establish the equivalence of several versions of the Nullstellensatz. 8.4.1 Theorem For any ﬁeld k and any positive integer n, the following statements are equivalent. (1) Maximal Ideal Theorem The maximal ideals of k[X1 , . . . , Xn ] are the ideals of the form (X1 − a1 , . . . , Xn − an ), a1 , . . . , an ∈ k. Thus maximal ideals correspond to points. (2) Weak Nullstellensatz If I is an ideal of k[X1 , . . . , Xn ] and V (I) = ∅, then I = k[X1 , . . . , Xn ]. Equivalently, if I is a proper ideal, then V (I) is not empty. (3) Nullstellensatz If I is an ideal of k[X1 , . . . , Xn ], then √ IV (I) = I.

page 9 of Chapter 8

(4) k is algebraically closed. Proof. (1) implies (2). Let I be a proper ideal, and let J be a maximal ideal containing I. By (8.1.3), part (4), V (J) ⊆ V (I), so it suﬃces to show that V (J) is not empty. By (1), J has the form (X1 − a1 , . . . , Xn − an ). But then a = (a1 , . . . , an ) ∈ V (J). [In fact V (J) = {a}.] (2) implies (3). This was done in Section 8.3. (3) implies (2). We use the fact that the radical of an ideal I is the intersection of all prime ideals containing I; see Section 8.3, Problem 2. Let I be a proper ideal of k[X1 , . . . √ , Xn ]. I is Then I is contained in a maximal, hence prime, ideal P . By the result just quoted, √ also contained in P , hence I is a proper ideal. By (3), IV (I) is a proper ideal. But if V (I) = ∅, then by (8.1.3) part (7), IV (I) = k[X1 , . . . , Xn ], a contradiction. (2) implies (1). If I is a maximal ideal, then by (2) there is a point a = (a1 , . . . , an ) ∈ V (I). Thus every f ∈ I vanishes at a, in other words, I ⊆ I({a}). But (X1 − a1 , . . . , Xn − an ) = I({a}); to see this, decompose f ∈ I({a}) as in the proof of (8.3.1). Therefore the maximal ideal I is contained in the maximal ideal (X1 − a1 , . . . , Xn − an , and it follows that I = (X1 − a1 , . . . , Xn − an ). (4) implies (1). Let I be a maximal ideal of k[X1 , . . . , Xn ], and let K = k[X1 , . . . , Xn ]/I, a ﬁeld containing an isomorphic copy of k via c → c + I, c ∈ k. By (8.3.9), K is a ﬁnite extension of k, so by (4), K = k. But then Xi + I = ai + I for some ai ∈ k, i = 1, . . . , n. Therefore Xi − ai is zero in k[X1 , . . . , Xn ]/I, in other words, Xi − ai ∈ I. Consequently, I ⊇ (X1 − a1 , . . . , Xn − an ), and we must have equality by (8.3.1). (1) implies (4). Let f be a nonconstant polynomial in k[X1 ] with no root in k. We can regard f is a polynomial in n variables with no root in An . Let I be a maximal ideal containing the proper ideal (f ). By (1), I is of the form (X1 − a1 , . . . , Xn − an ) = I({a}) for some a = (a1 , . . . , an ) ∈ An . Therefore f vanishes at a, a contradiction. ♣ 8.4.2 Corollary If the ideals I and J deﬁne the same variety and a polynomial g belongs to one of the ideals, then some power of g belongs to the other ideal. √ √ √ Proof. If V (I) = V (J), then by the Nullstellensatz, I = J. If g ∈ I ⊆ I, then g r ∈ J for some positive integer r. ♣ 8.4.3 Corollary The maps V → I(V ) and I → V (I) √ set up a one-to-one correspondence between varieties and radical ideals (deﬁned by I = I). √ Proof. By (8.1.3) part 6, V I(V ) = V . By the Nullstellensatz, IV (I) = I = I for radical ideals. It remains to prove that for any variety V , I(V ) is a radical ideal. If f r ∈ I(V ), then f r , hence f , vanishes on V , so f ∈ I(V ). ♣ 8.4.4 Corollary Let f1 , . . . , fr , g ∈ k[X1 , . . . , Xn ], and assume that g vanishes wherever the fi all vanish. Then there are polynomials h1 , . . . , hr ∈ k[X1 , . . . , Xn ] and a positive integer s such that g s = h1 f1 + · · · + hr fr . Proof. Let I be the ideal generated by f1 , . . . , fr . Then V (I) is the set of points at which all fi vanish, so that IV (I) is the√set of polynomials that vanish wherever all fi vanish. Thus g belongs to IV (I), which is I by the Nullstellensatz. Consequently, for some positive integer s, we have g s ∈ I, and the result follows. ♣ Problems For Section 8.4 1. Let f be a polynomial in k[X1 , . . . , Xn ], and assume that the factorization of f into irreducibles is f = f1n1 · · · frnr . Show that the decomposition of the variety V (f ) into irreducible subvarieties (Section 8.1, Problems 5 and 6) is given by V (f ) = ∪ri=1 V (fi ). 2. Under the hypothesis of Problem 1, show that IV (f ) = (f1 · · · fr ). 3. Show that there is a one-to-one correspondence between irreducible polynomials in

page 10 of Chapter 8

k[X1 , . . . , Xn ] and irreducible hypersurfaces (see (8.2.2))in An (k), if polynomials that diﬀer by a nonzero multiplicative constant are identiﬁed. 4. For any collection of subsets Xi of An , show that I(∪i Xi ) = ∩i I(Xi ). 5. Show that every radical ideal I of k[X1 , . . . , Xn ] is the intersection of ﬁnitely many prime ideals. 6. In Problem 5, show that the decomposition is unique, subject to the condition that the prime ideals P are minimal, that is, there is no prime ideal Q with I ⊆ Q ⊂ P . 7. Suppose that X is a variety in A2 , deﬁned by equations f1 (x, y) = · · · = fm (x, y) = 0, m ≥ 2. Let g be the greatest common divisor of the fi . If g is constant, show that X is a ﬁnite set (possibly empty). 8. Show that every variety in A2 except for A2 itself is the union of a ﬁnite set and an algebraic curve. 9. Give an example of two distinct irreducible polynomials in k[X, Y ] with the same zeroset, and explain why this cannot happen if k is algebraically closed. 10. Give an explicit example of the failure of a version of the Nullstellensatz in a nonalgebraically closed ﬁeld. 8.5 Localization 8.5.1 Geometric Motivation Suppose that V is an irreducible variety, so that I(V ) is a prime ideal. A polynomial g will belong to I(V ) if and only if it vanishes on V . If we are studying rational functions f /g in the neighborhood of a point x ∈ V , we must have g(x) = 0. It is very convenient to have every polynomial g ∈ / I(V ) available as a legal object, even though g may vanish at some points of V . The technical device that makes this possible is the construction of the ring of fractions S −1 R, the localization of R by S, where R = k[X1 , . . . , Xn ] and S is the multiplicative set R \ I(V ). We will now study the localization process in general. 8.5.2 Notation Recalling the setup of Section 2.8, let S be a multiplicative subset of the ring R, and S −1 R the ring of fractions of R by S. Let h be the natural homomorphism of R into S −1 R, given by h(a) = a/1. If X is any subset of R, deﬁne S −1 X = {x/s : x ∈ X, s ∈ S}. We will be especially interested in such a set when X is an ideal. If I and J are ideals of R, the product of I and J, denoted by IJ, is deﬁned (as in (7.6.1)) as the set of all ﬁnite sums i xi yi , xi ∈ I, yi ∈ J. It follows from the deﬁnition that IJ is an ideal. The sum of two ideals has already been deﬁned in (2.2.8). 8.5.3 Lemma If I is an ideal of R, then S −1 I is an ideal of S −1 R. If J is another ideal of R, then (i) S −1 (I + J) = S −1 I + S −1 J; (ii) S −1 (IJ) = (S −1 I)(S −1 J); (iii)S −1 (I ∩ J) = S −1 I ∩ S −1 J; (iv) S −1 I is a proper ideal iﬀ S ∩ I = ∅. Proof. The deﬁnition of addition and multiplication in S −1 R implies that S −1 I is an ideal, and that in (i), (ii) and (iii), the left side is contained in the right side. The reverse inclusions in (i) and (ii) follow from a b at + bs a b ab + = , = . s t st st st To prove (iii), let a/s = b/t where a ∈ I, b ∈ J, s, t ∈ S. There exists u ∈ S such that u(at − bs) = 0. Then a/s = uat/ust = ubs/ust ∈ S −1 (I ∩ J). Finally, if s ∈ S ∩ I then 1/1 = s/s ∈ S −1 I, so S −1 I = S −1 R. Conversely, if S −1 I = S R, then 1/1 = a/s for some a ∈ I, s ∈ S. There exists t ∈ S such that t(s − a) = 0, so at = st ∈ S ∩ I. ♣ −1

page 11 of Chapter 8

Ideals in S −1 R must be of a special form. 8.5.4 Lemma If J is an ideal of S −1 R and I = h−1 (J), then I is an ideal of R and S −1 I = J. Proof. I is an ideal by the basic properties of preimages of sets. Let a/s ∈ S −1 I, with a ∈ I and s ∈ S. Then a/1 ∈ J, so a/s = (a/1)(1/s) ∈ J. Conversely, let a/s ∈ J, with a ∈ R, s ∈ S. Then h(a) = a/1 = (a/s)(s/1) ∈ J, so a ∈ I and a/s ∈ S −1 I. ♣ Prime ideals yield sharper results. 8.5.5 Lemma If I is any ideal of R, then I ⊆ h−1 (S −1 I), with equality if I is prime and disjoint from S. Proof. If a ∈ I, then h(a) = a/1 ∈ S −1 I. Thus assume that I is prime and disjoint from S, and let a ∈ h−1 (S −1 I). Then h(a) = a/1 ∈ S −1 I, so a/1 = b/s for some b ∈ I, s ∈ S. There exists t ∈ S such that t(as − b) = 0. Thus ast = bt ∈ I, with st ∈ / I since S ∩ I = ∅. Since I is prime, we have a ∈ I. ♣ 8.5.6 Lemma If I is a prime ideal of R disjoint from S, then S −1 I is a prime ideal of S −1 R. Proof. By (8.5.3), part (iv), S −1 I is a proper ideal. Let (a/s)(b/t) = ab/st ∈ S −1 I, with a, b ∈ R, s, t ∈ S. Then ab/st = c/u for some c ∈ I, u ∈ S. There exists v ∈ S such that v(abu − cst) = 0. Thus abuv = cstv ∈ I, and uv ∈ / I because S ∩ I = ∅. Since I is prime, ab ∈ I, hence a ∈ I or b ∈ I. Therefore either a/s or b/t belongs to S −1 I. ♣ The sequence of lemmas can be assembled to give a precise conclusion. 8.5.7 Theorem There is a one-to-one correspondence between prime ideals P of R that are disjoint from S and prime ideals Q of S −1 R, given by P → S −1 P and Q → h−1 (Q). Proof. By (8.5.4), S −1 (h−1 (Q)) = Q, and by (8.5.5), h−1 (S −1 P ) = P . By (8.5.6), S −1 P is a prime ideal, and h−1 (Q) is a prime ideal by the basic properties of preimages of sets. If h−1 (Q) meets S, then by (8.5.3) part (iv), Q = S −1 (h−1 (Q)) = S −1 R, a contradiction. Thus the maps P → S −1 P and Q → h−1 (Q) are inverses of each other, and the result follows. ♣ 8.5.8 Deﬁnitions and Comments If P is a prime ideal of R, then S = R \ P is a multiplicative set. In this case, we write R(P ) for S −1 R, and call it the localization of R at P . (The usual notation is RP , but it’s easier to read without subscripts.) If I is an ideal of R, we write I(P ) for S −1 I. We are going to show that R(P ) is a local ring, that is, a ring with a unique maximal ideal. First we give some conditions equivalent to the deﬁnition of a local ring. 8.5.9 Proposition For a ring R, the following conditions are equivalent. (i) R is a local ring; (ii) There is a proper ideal I of R that contains all nonunits of R; (iii) The set of nonunits of R is an ideal. Proof. (i) implies (ii). If a is a nonunit, then (a) is a proper ideal, hence is contained in the unique maximal ideal I. (ii) implies (iii). If a and b are nonunits, so are a + b and ra. If not, then I contains a unit, so I = R, a contradiction. (iii) implies (i). If I is the ideal of nonunits, then I is maximal, because any larger ideal J would have to contain a unit, so that J = R. If H is any proper ideal, then H cannot contain a unit, so H ⊆ I. Therefore I is the unique maximal ideal. ♣

page 12 of Chapter 8

8.5.10 Theorem R(P ) is a local ring. Proof. Let Q be a maximal ideal of R(P ). Then Q is prime, so by (8.5.7), Q = I(P ) for some prime ideal I of R that is disjoint from S, in other words, contained in P . Thus Q = I(P ) ⊆ P (P ). If P (P ) = R(P ), then by (8.5.3) part (iv), P is not disjoint from S = R \ P , which is impossible. Therefore P (P ) is a proper ideal containing every maximal ideal, so it must be the unique maximal ideal. ♣ If R is an integral domain and S is the set of all nonzero elements of R, then S −1 R is the quotient ﬁeld of R. In this case, S −1 R is a local ring, because any ﬁeld is a local ring. ({0} is the unique maximal ideal.) Alternatively, we can appeal to (8.5.10) with P = {0}. 8.5.11 Localization of Modules If M is an R-module and S a multiplicative subset of R, we can essentially repeat the construction of Section 2.8 to form the localization S −1 M of M by S, and thereby divide elements of M by elements of S. If x, y ∈ M and s, t ∈ S, we call (x, s) and (y, t) equivalent if for some u ∈ S, u(tx − sy) = 0. The equivalence class of (x, s) is denoted by x/s, and addition is deﬁned by x y tx + sy + = . s t st If a/s ∈ S −1 R and x/t ∈ S −1 M , we deﬁne ax ax = . st st In this way, S −1 M becomes an S −1 R-module. Exactly as in (8.5.3), if M and N are submodules of a module L, then S −1 (M + N ) = S −1 M + S −1 N and S −1 (M ∩ N ) = S −1 M ∩ S −1 N. Further properties will be given in the exercises. Problems For Section 8.5 1. Let M be a maximal ideal of R, and assume that for every x ∈ M, 1 + x is a unit. Show that R is a local ring (with maximal ideal M ). [Show that if x ∈ / M , then x is a unit, and apply (8.5.9).] 2. Show that if p is prime and n is a positive integer, then Zpn is a local ring with maximal ideal (p). 3. Let R be the ring of all n by n matrices with coeﬃcients in a ﬁeld F . If A is a nonzero element of R and 1 is the identity matrix, is {1, A, A2 , . . .} always a multiplicative set? Let S be a multiplicative subset of the ring R. We are going to construct a mapping from R-modules to S −1 R-modules, and another mapping from R-module homomorphisms to S −1 R-module homomorphisms, as follows. If M is an R-module, we let M → S −1 M . If f : M → N is an R-module homomorphism, we deﬁne S −1 f : S −1 M → S −1 N by x f (x) → . s s Since f is a homomorphism, so is S −1 f . 4. If g : N → L and composition of functions is written as a product, show that S −1 (gf ) = S −1 (g)S −1 (f ), and if 1M is the identity mapping on M , then S −1 (1M ) = 1S −1 M . We say that S −1 is a functor from the category of R-modules to the category of S −1 R-modules. This terminology will be explained in great detail in Chapter 10. 5. If f g M → N → L

page 13 of Chapter 8

is an exact sequence, show that S

−1

M

S −1 f →

S

−1

N

S −1 g →

S −1 L

is exact. We say that S −1 is an exact functor. Again, we will study this idea in Chapter 10. 6. Let R be the ring of rational functions f /g with f, g ∈ k[X1 , . . . , Xn ] and g(a) = 0, where a = (a1 , . . . , an ) is a ﬁxed point in An . Show that R is a local ring, and identify the unique maximal ideal. 7. If M is an R-module and S is a multiplicative subset of R, denote S −1 M by MS . If N is a submodule of M , show that (M/N )S ∼ = MS /NS . 8.6 Primary Decomposition We have seen that every radical ideal in k[X1 , . . . , Xn ] can be expressed as an intersection of ﬁnitely many prime ideals (Section 8.4, Problem 5). A natural question is whether a similar result holds for arbitrary ideals. The answer is yes if we generalize from prime to primary ideals. 8.6.1 Deﬁnitions and Comments The ideal Q in the ring R is primary if Q is proper and whenever a product ab belongs to Q, either a ∈ Q or bn ∈ Q for some positive integer √ n. [The condition on b is equivalent to b ∈ Q.] An equivalent statement is that R/Q = 0 and whenever (a + Q)(b + Q) = 0 in R/Q, either a + Q = 0 or (b + Q)n = 0 for some positive integer n. This says that if b + Q is a zero-divisor in R/Q, then it is nilpotent, that is, some power of b + Q is 0. It is primary. Also, if Q is primary, √ follows from the deﬁnition that every prime ideal √ then Q is the smallest prime ideal containing Q. [Since Q is the intersection of all prime √ ideals containing Q (Section 8.3, Problem 2), it suﬃces to show that Q is prime. √ But if an bn ∈ Q, then an ∈ Q or bnm ∈ Q for some m, so either a or b must belong to √ Q. Note also that since Q is proper, it is contained in a maximal, hence prime, ideal, so Q is also proper.] √ If Q is primary and Q = P , we say that Q is P -primary. 8.6.2 Examples 1. In Z, the primary ideals are {0} and (pr ), where p is prime. In Z6 , 2 and 3 are zerodivisors that are not nilpotent, and a similar situation will occur in Zm whenever more than one prime appears in the factorization of m. 2. Let R = k[X, Y ] where k is any ﬁeld, and take Q = (X, Y 3 ), the ideal generated by X and Y 3 . This is a nice example of analysis in quotient rings. We are essentially setting X and Y 3 equal to zero, and this collapses the ring R down to polynomials a0 + a1 Y + a2 Y 2 , with the ai ∈ k and arithmetic mod Y 3 . Formally, R/Q is isomorphic to k[Y ]/(Y 3 ). The zero-divisors in R/Q are of the form cY + dY 2 , c ∈ k, and they are nilpotent. Thus Q is primary. If f ∈ R, then the only way for √ f not to belong to the radical of Q is for the constant term of f to be nonzero. Thus Q = (X, Y ), a maximal ideal by (8.3.1). Now we claim that Q cannot be a power of a prime ideal; this will be a consequence of the next result. √ 8.6.3 Lemma If P is a prime ideal, then for every positive integer n, P n = P . √ n n n n Proof. √ Since P is a prime ideal containing P , P ⊆ P . If x ∈ P , then x ∈ P , so x ∈ P n. ♣ Returning to Example 2 of (8.6.2), if Q = (X, Y 3 ) is a prime power P n , then its radical is P , so P must be (X, Y ). But X ∈ Q and X ∈ / P n , n ≥ 2; since Y belongs to P but not Q, we have reached a contradiction.

page 14 of Chapter 8

After a preliminary deﬁnition, we will give a convenient suﬃcient condition for an ideal to be primary. 8.6.4 Deﬁnition The nilradical N (R) of a ring R is the set of nilpotent elements of R, that is, {x ∈ R : xn = 0 for some positive integer n}. Thus N (R) is the radical of the zero ideal, which is the intersection of all prime ideals of R. 8.6.5 Proposition If the radical of the ideal Q is maximal, then Q is primary. √ Proof. Since Q is maximal, it must be the only prime ideal containing Q. By the correspondence theorem and the fact that the preimage of a prime ideal is a prime ideal (cf. (8.5.7)), R/Q has exactly one prime ideal, which must coincide with N (R/Q). Any element of R/Q that is not a unit generates a proper ideal, which is contained in a maximal ideal, which again must be N (R/Q). Thus every element of R/Q is either a unit or nilpotent. Since a zero-divisor cannot be a unit, every zero-divisor of R/Q is nilpotent, so Q is primary. ♣ 8.6.6 Corollary If M is a maximal ideal, then M n is M -primary for all n = 1, 2, . . .. Proof. By (8.6.3), the radical of M n is M , and the result follows from (8.6.5). ♣ Here is another useful property. 8.6.7 Proposition If Q is a ﬁnite intersection of P -primary ideals Qi , i = 1, . . . , n, then Q is P -primary. Proof. First note that the radical of a ﬁnite intersection of ideals is the intersection of the radicals (see Problem 1). It follows that the radical of Q is P , and it remains to show that Q is primary. If ab / Q, then for some i we have a ∈ / Qi . Since Qi is P -primary, √∈ Q but a ∈ b belongs to P = Qi . But then some power of b belongs to Q. ♣ We are going to show that in a Noetherian ring, every proper ideal I has a primary decomposition, that is, I can be expressed as a ﬁnite intersection of primary ideals. 8.6.8 Lemma Call an ideal I irreducible if for any ideals J and K, I = J ∩ K implies that I = J or I = K. If R is Noetherian, then every ideal of R is a ﬁnite intersection of irreducible ideals. Proof. Suppose that the collection S of all ideals that cannot be so expressed is nonempty. Since R is Noetherian, S has a maximal element I, necessarily reducible. Let I = J ∩ K, where I is properly contained in both J and K. By maximality of I, the ideals J and K are ﬁnite intersections of irreducible ideals, and consequently so is I, contradicting I ∈ S. ♣ If we can show that every irreducible proper ideal is primary, we then have the desired primary decomposition. Let us focus on the chain of reasoning we will follow. If I is an irreducible proper ideal of R, then by the correspondence theorem, 0 is an irreducible ideal of the Noetherian ring R/I. If we can show that 0 is primary in R/I, then again by the correspondence theorem, I is primary in R. 8.6.9 Primary Decomposition Theorem Every proper ideal in a Noetherian ring R has a primary decomposition. (We can drop the word “proper” if we regard R as the intersection of the empty collection of primary ideals.) Proof. By the above discussion, it suﬃces to show that if 0 is an irreducible ideal of R, then it is primary. Let ab = 0 with a = 0. Since R is Noetherian, the sequence of annihilators ann b ⊆ ann b2 ⊆ ann b3 ⊆ · · · stabilizes, so ann bn = ann bn+1 for some n. If we can show that (a) ∩ (bn ) = 0

page 15 of Chapter 8

we are ﬁnished, because a = 0 and the zero ideal is irreducible (by hypothesis). Thus let x = ca = dbn for some c, d ∈ R. Then bx = cab = dbn+1 = 0 (because ab = 0), so d annihilates bn+1 , hence d annihilates bn . Thus x = dbn = 0. ♣ Problems For Section 8.6 1. If I1 , . . . , In are arbitrary ideals, show that n n

√

Ii = I i. i=1

i=1

2. Let I be the ideal (XY − Z 2 ) in k[X, Y, Z], where k is any ﬁeld, and let R = k[X, Y, Z]/I. If P is the ideal (X + I, Z + I), show that P is prime. 3. Continuing Problem 2, show that P 2 , whose radical is prime by (8.6.3) and which is a power of a prime, is nevertheless not primary. 4. Let R = k[X, Y ], and let P1 = (X), P2 = (X, Y ), Q = (X 2 , Y ). Show that P1 is prime and P22 and Q are P2 -primary. 5. Continuing Problem 4, let I = (X 2 , XY ). Show that P1 ∩ P22 and P1 ∩ Q are both primary decompositions of I. Notice that the radicals of the components of the primary decomposition (referred to √ as the √ primes associated with I) are P1 and P in both cases. [P is prime, so P = P 2 1 1; 1 √ P2 ⊆ Q and P2 is maximal, so P2 = Q; ] Uniqueness questions involving primary decompositions are treated in detail in textbooks on commutative algebra. 6. We have seen in Problem 5 of Section 8.4 that every radical ideal in R = k[X1 , . . . , Xn ] is the intersection of ﬁnitely many prime ideals. Show that this result holds in an arbitrary Noetherian ring R. 7. Let R = k[X, Y ] and let In be the ideal (X 3 , XY, Y n ). Show that for every positive integer n, In is a primary ideal of R. 8.7 Tensor Product of Modules Over a Commutative Ring 8.7.1 Motivation In many areas of algebra and its applications, it is useful to multiply, in a sensible way, an element x of an R-module M by an element y of an R-module N . In group representation theory, M and N are free modules, in fact ﬁnite-dimensional vector spaces, with bases {xi } and {yj }. Thus if we specify that multiplication is linear in each variable, then we need only specify products of xi and yj . We require that the these products, to be denoted by xi ⊗ yj , form a basis for a new R-module T . If f : R → S is a ring homomorphism and M is an S-module, then M becomes an R-module via rx = f (r)x, r ∈ R, x ∈ M . This is known as restriction of scalars. In algebraic topology and algebraic number theory, it is often desirable to reverse this process. If M is an R-module, we want to extend the given multiplication rx, r ∈ R, x ∈ M , to multiplication of an arbitrary s ∈ S by x ∈ M . This is known as extension of scalars, and it becomes possible with the aid of the tensor product construction. The tensor product arises in algebraic geometry in the following way. Let M be the coordinate ring of a variety V in aﬃne space Am , in other words, M is the set of all polynomial functions from V to the base ﬁeld k. Let N be the coordinate ring of the variety W in An . Then the cartesian product V × W is a variety in Am+n , and its coordinate ring turns out to be the tensor product of M and N . Let’s return to the ﬁrst example above, where M and N are free modules with bases {xi } and {yj }. Suppose that f is a bilinear map from M × N to an R-module P . (In other words, f is R-linear in each variable.) Information about f can be completely encoded into a function g of one variable, where g is an R-module homomorphism from T to P . We take g(xi ⊗ yj ) = f (xi , yj ) and extend by linearity. Thus f is the composition of the bilinear map h from M × N to T speciﬁed by (xi , yj ) → xi ⊗ yj , followed by g. To summarize:

page 16 of Chapter 8

Every bilinear mapping on M × N can be factored through T. The R-module T is called the tensor product of M and N , and we write T = M ⊗R N . We are going to construct a tensor product of arbitrary modules over a commutative ring, and sketch the generalization to noncommutative rings. 8.7.2 Deﬁnitions and Comments Let M and N be arbitrary R-modules, and let F be a free R-module with basis M ×N . Let G be the submodule of F generated by the “relations” (x + x , y) − (x, y) − (x , y); (rx, y) − r(x, y);

(x, y + y ) − (x, y) − (x, y ); (x, ry) − r(x, y)

where x, x ∈ M, y, y ∈ N, r ∈ R. Deﬁne the tensor product of M and N (over R) as T = M ⊗R N = F/G and denote the element (x, y) + G of T by x ⊗ y. Thus the general element of T is a ﬁnite sum of the form t= xi ⊗ yi (1) i

with xi ∈ M and yi ∈ N . It is important to note that the representation (1) is not necessarily unique. The relations force x ⊗ y to be linear in each variable, so that x ⊗ (y + y ) = x ⊗ y + x ⊗ y , (x + x ) ⊗ y = x ⊗ y + x ⊗ y,

(2)

r(x ⊗ y) = rx ⊗ y = x ⊗ ry.

(3)

See Problem 1 for details. Now if f is a bilinear mapping from M × N to the R-module P , then f extends uniquely to a homomorphism from F to P , also called f . Bilinearity means that the kernel of f contains G, so by the factor theorem, there is a unique R-homomorphism g : T → P such that g(x ⊗ y) = f (x, y) for all x ∈ M, y ∈ N . As in (8.7.1), if we compose the bilinear map h : (x, y) → x ⊗ y with g, the result is f . Again, we say that Every bilinear mapping on M × N can be factored through T . We have emphasized this sentence, known as a universal mapping property (abbreviated UMP), because along with equations (1), (2) and (3), it indicates how the tensor product is applied in practice. The detailed construction we have just gone through can now be forgotten. In fact any two R-modules that satisfy the universal mapping property are isomorphic. The precise statement and proof of this result will be developed in the exercises. In a similar fashion, using multilinear rather than bilinear maps, we can deﬁne the tensor product of any ﬁnite number of R-modules. [In physics and diﬀerential geometry, a tensor is a multilinear map on a product M1 × · · · × Mr , where each Mi is either a ﬁnitedimensional vector space V or its dual space V ∗ . This suggests where the terminology “tensor product” comes from.] In the discussion to follow, M , N and P are R-modules. The ring R is assumed ﬁxed, and we will usually write ⊗ rather than ⊗R . 8.7.3 Proposition M ⊗ N ∼ = N ⊗ M. Proof. Deﬁne a bilinear mapping f : M × N → N ⊗ M by f (x, y) = y ⊗ x. By the UMP, there is a homomorphism g : M ⊗ N → N ⊗ M such that g(x ⊗ y) = y ⊗ x. Similarly, there is a homomorphism g : N ⊗ M → M ⊗ N with g (y ⊗ x) = x ⊗ y. Thus g is an isomorphism (with inverse g ). ♣

page 17 of Chapter 8

8.7.4 Proposition M ⊗ (N ⊗ P ) ∼ = (M ⊗ N ) ⊗ P . Proof. Deﬁne f : M × N × P → (M ⊗ N ) ⊗ P by f (x, y, z) = (x ⊗ y) ⊗ z. The UMP produces g : M × (N ⊗ P ) → (M ⊗ N ) ⊗ P with g((x, (y ⊗ z))) = (x ⊗ y) ⊗ z. [We are applying the UMP for each ﬁxed x ∈ M , and assembling the maps to produce g.] Since g is bilinear (by Equations (2) and (3)), the UMP yields h : M ⊗ (N ⊗ P ) → (M ⊗ N ) ⊗ P with h(x ⊗ (y ⊗ z)) = (x ⊗ y) ⊗ z. Exactly as in (8.7.3), we can construct the inverse of h, so h is the desired isomorphism. ♣ 8.7.5 Proposition M ⊗ (N ⊕ P ) ∼ = (M ⊗ N ) ⊕ (M ⊗ P ). Proof. Let f be an arbitrary bilinear mapping from M × (N ⊕ P ) to Q. If x ∈ M, y ∈ N, z ∈ P , then f (x, y + z) = f (x, y) + f (x, z). The UMP gives homomorphisms g1 : M ⊗ N → Q and g2 : M ⊗ P → Q such that g1 (x ⊗ y) = f (x, y) and g2 (x ⊗ z) = f (x, z). The maps g1 and g2 combine to give g : (M ⊗ N ) ⊕ (M ⊗ P ) → Q such that g((x ⊗ y) + (x ⊗ z)) = g1 (x ⊗ y) + g2 (x ⊗ z). In particular, with x = x, g((x ⊗ y) + (x ⊗ z)) = f (x, y + z), so if h : M × (N ⊕ P ) → M ⊗ (N ⊕ P ) is deﬁned by h(x, y + z) = (x ⊗ y) + (x ⊗ z), then f = gh. Thus (M ⊗ N ) ⊕ (M ⊗ P ) satisﬁes the universal mapping property, hence must be isomorphic to the tensor product. ♣ 8.7.6 Proposition Regarding R as a module over itself, R ⊗R M ∼ = M. Proof. The map (r, x) → rx of R × M → M is bilinear, so there is a homomorphism g : R ⊗ M → M such that g(r ⊗ x) = rx. Deﬁne h : M → R ⊗ M by h(x) = 1 ⊗ x. Then h(rx) = 1 ⊗ rx = r1 ⊗ x = r ⊗ x. Thus g is an isomorphism (with inverse h). ♣ 8.7.7 Corollary Let Rm be the direct sum of m copies of R, and M m the direct sum of m copies of M . Then Rm ⊗ M ∼ = M m. Proof. By (8.7.5), Rm ⊗ M is isomorphic to the direct sum of m copies of R ⊗ M , which is isomorphic to M m by (8.7.6). ♣ ∼ Rmn . Moreover, if {x1 , . . . , xm } is a basis for Rm and 8.7.8 Proposition Rm ⊗ Rn = n {y1 , . . . , yn } is a basis for R , then {xi ⊗ yj , i = 1, . . . , m, j = 1, . . . , n} is a basis for Rmn . Proof. This follows from the discussion in (8.7.1). [The ﬁrst assertion can also be proved by taking M = Rn in (8.7.7).] ♣ 8.7.9 Tensor Product of Homomorphisms Let f1 : M1 → N1 and f2 : M2 → N2 be R-module homomorphisms. The map (x1 , x2 ) → f1 (x1 ) ⊗ f2 (x2 ) of M1 × M2 into N1 ⊗ N2 is bilinear, and induces a unique f : M1 ⊗ M2 → N1 ⊗ N2 such that f (x1 ⊗ x2 ) = f1 (x1 ) ⊗ f2 (x2 ), x1 ∈ M1 , x2 ∈ M2 . We write f = f1 ⊗ f2 , and call it the tensor product of f1 and f2 . Similarly, if g1 : N1 → P1 and g2 : N2 → P2 , then we can compose g1 ⊗ g2 with f1 ⊗ f2 , and (g1 ⊗ g2 )(f1 ⊗ f2 )(x1 ⊗ x2 ) = g1 f1 (x1 ) ⊗ g2 f2 (x2 ), hence (g1 ⊗ g2 ) ◦ (f1 ⊗ f2 ) = (g1 ◦ f1 ) ⊗ (g2 ◦ f2 ).

page 18 of Chapter 8

When M1 = N1 = V , a free R-module of rank m, and M2 = N2 = W , a free R-module of rank n, there is a very concrete interpretation of the tensor product of the endomorphisms f : V → V and g : W → W . If f is represented by the matrix A and g by the matrix B, then the action of f and g on basis elements is given by f (vj ) = aij vi , g(wl ) = bkl wk i

k

where i and j range from 1 to m, and k and l range from 1 to n. Thus (f ⊗ g)(vj ⊗ wl ) = f (vj ) ⊗ g(wl ) = aij bkl (vi ⊗ wk ). i,k

The mn by mn matrix representing the endomorphism f ⊗ g : V ⊗ W → V ⊗ W is denoted by A ⊗ B and called the tensor product or Kronecker product of A and B. It is given by 

a11 B . A ⊗ B =  .. am1 B

 a1m B ..  . .

a12 B

···

am2 B

· · · amm B

The ordering of the basis of V ⊗ W is v1 ⊗ w1 , . . . , v1 ⊗ wn , . . . , vm ⊗ w1 , . . . , vm ⊗ wn . To determine the column of A ⊗ B corresponding to vj ⊗ wl , locate the aij B block (i = 1, . . . , m; j ﬁxed) and proceed to column l of B. As we move down this column, the indices i and k vary according to the above ordering of basis elements. If this road map is not clear, perhaps writing out the entire matrix for m = 2 and n = 3 will help. Problems For Section 8.7 1. Verify Equations (2) and (3) of (8.7.2). 2. If m and n are relatively prime, show that Zm ⊗Z Zn = 0. 3. If A is a ﬁnite abelian group and Q is the additive group of rationals, show that A⊗Z Q = 0. Generalize to a wider class of abelian groups A. 4. The deﬁnition of M ⊗R N via a universal mapping property is as follows. The tensor product is an R-module T along with a bilinear map h : M × N → T such that given any bilinear map f : M × N → P , there is a unique R-homomorphism g : T → P such that f = gh. See the diagram below. h

M×N

T g

f

P Now suppose that another R-module T , along with a bilinear mapping h : M ×N → T , satisﬁes the universal mapping property. Using the above diagram with P = T and f replaced by h , we get a unique homomorphism g : T → T such that h = gh. Reversing the roles of T and T , we get g : T → T such that h = g h . Show that T and T are isomorphic. 5. Consider the element n ⊗ x in Z ⊗ Zn , where x is any element of Zn and we are tensoring over Z, i.e., R = Z. Show that n ⊗ x = 0. 6. Continuing Problem 5, take x = 0 and regard n ⊗ x as an element of nZ ⊗ Zn rather than Z ⊗ Zn . Show that n ⊗ x = 0. 7. Let M, N, M , N be arbitrary R-modules, where R is a commutative ring. Show that the

page 19 of Chapter 8

tensor product of homomorphisms induces a linear map from HomR (M, M )⊗R HomR (N, N ) to HomR (M ⊗R N, M ⊗R N ). 8. Let M be a free R-module of rank m, and N a free R-module of rank n. Show that there is an R-module isomorphism of EndR (M )⊗R EndR (N ) and EndR (M ⊗ N ). 8.8 General Tensor Products We now consider tensor products of modules over noncommutative rings. A natural question is “Why not simply repeat the construction of (8.7.2) for an arbitrary ring R?”. But this construction forces rx ⊗ sy = r(x ⊗ sy) = rs(x ⊗ y) and rx ⊗ sy = s(rx ⊗ y) = sr(x ⊗ y) which cannot hold in general if R is noncommutative. A solution is to modify the construction so that the tensor product T is only an abelian group. Later we can investigate conditions under which T has a module structure as well. 8.8.1 Deﬁnitions and Comments Let M be a right R-module and N a left R-module. (We often abbreviate this as MR and R N .) Let f : M ×N → P , where P is an abelian group. The map f is biadditive if it is additive in each variable, that is, f (x+x , y) = f (x, y)+f (x , y) and f (x, y + y ) = f (x, y) + f (x, y ) for all x, x ∈ M, y, y ∈ N . The map f is R-balanced if f (xr, y) = f (x, ry) for all x ∈ M, y ∈ N, r ∈ R. As before, the key idea is the universal mapping property: Every biadditive, R-balanced map can be factored through the tensor product. 8.8.2 Construction of the General Tensor Product If MR and R N , let F be the free abelian group with basis M × N . Let G be the subgroup of R generated by the relations (x + x , y) − (x, y) − (x , y); (x, y + y ) − (x, y) − (x, y ); (xr, y) − (x, ry) where x, x ∈ M, y, y ∈ N, r ∈ R. Deﬁne the tensor product of M and N over R as T = M ⊗R N = F/G and denote the element (x, y) + G of T by x ⊗ y. Thus the general element of T is a ﬁnite sum of the form t= xi ⊗ yi . (1) i

The relations force the map h : (x, y) → x ⊗ y of M × N into T to be biadditive and R-balanced, so that x ⊗ (y + y ) = x ⊗ y + x ⊗ y , (x + x ) ⊗ y = x ⊗ y + x ⊗ y,

(2)

xr ⊗ y = x ⊗ ry.

(3)

If f is a biadditive, R-balanced mapping from M × N to the abelian group P , then f extends uniquely to an abelian group homomorphism from F to P , also called f . Since f is biadditive and R-balanced, the kernel of f contains G, so by the factor theorem there is a unique abelian group homomorphism g : T → P such that g(x ⊗ y) = f (x, y) for all x ∈ M, y ∈ N . Consequently, gh = f and we have the universal mapping property: Every biadditive, R-balanced mapping on M × N can be factored through T .

page 20 of Chapter 8

As before, any two abelian groups that satisfy the universal mapping property are isomorphic. 8.8.3 Bimodules Let R and S be arbitrary rings. We say that M is an S − R bimodule if M is both a left S-module and a right R-module, and in addition a compatibility condition is satisﬁed: (sx)r = s(xr) for all s ∈ S, r ∈ R. We often abbreviate this as S MR . If f : R → S is a ring homomorphism, then S is a left S-module, and also a right R-module by restriction of scalars, as in (8.7.1). The compatibility condition is satisﬁed: (sx)r = sxf (r) = s(xr). Therefore S is an S − R bimodule. 8.8.4 Proposition If S MR and R NT , then M ⊗R N is an S − T bimodule. Proof. Fix s ∈ S. The map (x, y) → sx ⊗ y of M × N into M ⊗R N is biadditive and Rbalanced. The latter holds because by the compatibility condition in the bimodule property of M , along with (3) of (8.8.2), s(xr) ⊗ y = (sx)r ⊗ y = sx ⊗ ry. Thus there is an abelian group endomorphism on M ⊗R N such that x ⊗ y → sx ⊗ y, and we use this to deﬁne scalar multiplication on the left by s. A symmetrical argument yields scalar multiplication on the right by t. To check the compatibility condition, [s(x ⊗ y)]t = (sx ⊗ y)t = sx ⊗ yt = s(x ⊗ yt) = s[(x ⊗ y)t]. ♣ 8.8.5 Corollary If S MR and R N , then M ⊗R N is a left S-module. If MR and R NT , then M ⊗R N is a right T -module. Proof. The point is that every module is, in particular, an abelian group, hence a Z-module. Thus for the ﬁrst statement, take T = Z in (8.8.4), and for the second statement, take S = Z. ♣ 8.8.6 Extensions As in Section 8.7, we can deﬁne the tensor product of any ﬁnite number of modules using multiadditive maps (additive in each variable) that are balanced. For example, suppose that MR , R NS and S P . If f : M × N × P → G, where G is an abelian group, the condition of balance is f (xr, y, z) = f (x, ry, z) and f (x, ys, z) = f (x, y, sz) for all x ∈ M, y ∈ N, z ∈ P, r ∈ R, s ∈ S. An argument similar to the proof of (8.7.4) shows that (a) M ⊗R N ⊗S P ∼ = (M ⊗R N ) ⊗S P ∼ = M ⊗R (N ⊗S P ). If M is a right R-module, and N and P are left R-modules, then (b) M ⊗R (N ⊕ P ) ∼ = (M ⊗R N ) ⊕ (M ⊗R P ). This is proved as in (8.7.5), in fact the result can be extended to the direct sum of an arbitrary (not necessarily ﬁnite) number of left R-modules. If M is a left R-module, then exactly as in (8.7.6) and (8.7.7), we have (c) R ⊗R M ∼ = M and m (d) R ⊗ M ∼ = M m. Let M1 and M2 be right R-modules, and let N1 and N2 be left R-modules. If f1 : M1 → N1 and f2 : M2 → N2 are R-module homomorphisms, the tensor product f1 ⊗ f2 can be deﬁned exactly as in (8.7.9). As before, the key property is (f1 ⊗ f2 )(x1 ⊗ x2 ) = f1 (x1 ) ⊗ f2 (x2 )

page 21 of Chapter 8

for all x1 ∈ M1 , x2 ∈ M2 . 8.8.7 Tensor Product of Algebras If A and B are algebras over the commutative ring R, then the tensor product A ⊗R B becomes an R-algebra if we deﬁne multiplication appropriately. Consider the map of A × B × A × B into A ⊗R B given by (a, b, a , b ) → aa ⊗ bb , a, a ∈ A, b, b ∈ B. The map is 4-linear, so it factors through the tensor product to give an R-module homomorphism g : A ⊗ B ⊗ A ⊗ B → A ⊗ B such that g(a ⊗ b ⊗ a ⊗ b ) = aa ⊗ bb . Now let h : (A ⊗ B) × (A ⊗ B) → A ⊗ B ⊗ A ⊗ B be the bilinear map given by h(u, v) = u ⊗ v. If we apply h followed by g, the result is a bilinear map f : (A ⊗ B) × (A ⊗ B) → A ⊗ B with f (a ⊗ b, a ⊗ b ) = aa ⊗ bb , and this deﬁnes our multiplication (a ⊗ b)(a ⊗ b ) on A ⊗ B. The multiplicative identity is 1A ⊗1B , and the distributive laws can be checked routinely. Thus A⊗R B is a ring that is also an R-module. To check the compatibility condition, note that if r ∈ R, a, a ∈ A, b, b ∈ B, then r[(a ⊗ b)(a ⊗ b )] = [r(a ⊗ b)](a ⊗ b ) = (a ⊗ b)[r(a ⊗ b )]; all three of these expressions coincide with raa ⊗ bb = aa ⊗ rbb . Problems For Section 8.8 We will use the tensor product to deﬁne the exterior algebra of an R-module M , where R is a commutative ring. If p is a positive integer, we form the tensor product M ⊗R · · ·⊗R M of M with itself p times, denoted by M ⊗p . Let N be the submodule of M ⊗p generated by those elements x1 ⊗ · · · ⊗ xp , with xi ∈ M for all i, such that xi = xj for some i = j. The pth exterior power of M is deﬁned as Λp M = M ⊗p /N. In most applications, M is a free R-module with a ﬁnite basis x1 , . . . , xn (with 1 ≤ p ≤ n), and we will only consider this case. To simplify the notation, we write the element a ⊗ b ⊗ · · · ⊗ c + N of Λp M as ab · · · c. (The usual notation is a ∧ b ∧ · · · ∧ c.) 1. Let y1 , . . . , yp ∈ M . Show that if yi and yj are interchanged in the product y1 · · · yp , then the product is multiplied by -1. 2. Show that the products xi1 · · · xip , where i1 < · · · < ip , span Λp M . 3. Let f : M p → Q be a multilinear map from M p to the R-module Q, and assume that f is alternating, that is, f (m1 , . . . , mp ) = 0 if mi = mj for some i = j. Show that f can be factored through Λp M , in other words, there is a unique R-homomorphism g : Λp M → Q such that g(y1 · · · yp ) = f (y1 , . . . , yp ). n Let yi = j=1 aij xj , i = 1, . . . , n. Since {x1 , . . . , xn } is a basis for M , yi can be identiﬁed with row i of A. By the basic properties of determinants, the map f (y1 , . . . , yn ) = det A is multilinear and alternating, and f (x1 , . . . , xn ) = 1, the determinant of the identity matrix. 4. Show that x1 · · · xn = 0 in Λn M , and that {x1 · · · xn } is a basis for Λn M .

page 22 of Chapter 8

Let I = {i1 , · · · , ip }, where i1 < · · · < ip , and write the product xi1 · · · xip as xI . Let J be the complementary set of indices. (For example, if n = 5, p = 3, and I = {1, 2, 4}, then J = {3, 5}.) Any equation involving xI ∈ Λp M can be multiplied by xJ to produce a valid equation in Λn M . 5. Show that the products xI of Problem 2 are linearly independent, so that Λp M is a free R-module of rank np . Roughly speaking, the exterior algebra of M consists of the Λp M for all p. By construction, Λ1 M = M and Λp M = 0 for p > n, since some index must repeat in any element of Λp M . By convention, we take Λ0 M = R. Formally, the exterior powers are assembled into a graded R-algebra A0 ⊕ A1 ⊕ A2 ⊕ · · · where Ap = Λp M . Multiplication is deﬁned as in the discussion after Problem 4, that is, if y1 · · · yp ∈ Ap and z1 · · · zq ∈ Aq , then the exterior product y1 · · · yp z1 · · · zq belongs to Ap+q . A ring R is said to be graded if, as an abelian group, it is the direct sum of subgroups Rn , n = 0, 1, 2, . . ., with Rm Rn ⊆ Rn+m for all m, n ≥ 0. [Example: R = k[X1 , . . . , Xn ], Rn = all homogeneous polynomials of degree n.] By deﬁnition, R0 is a subring of R (because R0 R0 ⊆ R0 ), and each Rn is a module over R0 (because R0 Rn ⊆ Rn ). 6. Suppose that the ideal I = ⊕n≥1 Rn is generated over R by ﬁnitely many elements x1 , . . . , xr , with xi ∈ Rni . Show that Rn ⊆ S = R0 [x1 , . . . , xr ] for all n = 0, 1, . . ., so that R = S. 7. Show that R is a Noetherian ring if and only if R0 is Noetherian and R is a ﬁnitely generated R0 -algebra.

page 1 of Chapter 9

CHAPTER 9 INTRODUCING NONCOMMUTATIVE ALGEBRA We will discuss noncommutative rings and their modules, concentrating on two fundamental results, the Wedderburn structure theorem and Maschke’s theorem. Further insight into the structure of rings will be provided by the Jacobson radical. 9.1 Semisimple Modules A vector space is the direct sum of one-dimensional subspaces (each subspace consists of scalar multiples of a basis vector). A one-dimensional space is simple in the sense that it does not have a nontrivial proper subspace. Thus any vector space is a direct sum of simple subspaces. We examine those modules which behave in a similar manner. 9.1.1 Deﬁnition An R-module M is simple if M = 0 and the only submodules of M are 0 and M . 9.1.2 Theorem Let M be a nonzero R-module. The following conditions are equivalent, and a module satisfying them is said to be semisimple or completely reducible. (a) M is a sum of simple modules; (b) M is a direct sum of simple modules; (c) If N is a submodule of M , then N is a direct summand of M , that is, there is a submodule N of M such that M = N ⊕ N . Proof.

(a) implies (b). Let M be the sum of simple modules Mi , i ∈ I. If J ⊆ I, denote j∈J Mj by M (J). By Zorn’s lemma, there is a maximal subset J of I such that the sum deﬁning N = M (J) is direct. We will show that M = N . First assume that i ∈ / J. Then N ∩ Mi is a submodule of the simple module Mi , so it must be either 0 or Mi . If N ∩ Mi = 0, then M (J ∪ {i}) is direct, contradicting maximality of J. Thus N ∩ Mi = Mi , so Mi ⊆ N . But if i ∈ J, then Mi ⊆ N by deﬁnition of N . Therefore Mi ⊆ N for all i, and since M is the sum of all the Mi , we have M = N . (b) implies (c). This is essentially the same as (a) implies (b). Let N be a submodule of M , where M is the direct sum of simple modules Mi , i ∈ I. Let J be a maximal subset of I such that the sum N + M (J) is direct. If i ∈ / J then exactly as before, Mi ∩ (N ⊕ M (J)) = Mi , so Mi ⊆ N ⊕ M (J). This holds for i ∈ J as well, by deﬁnition of M (J). It follows that M = N ⊕ M (J). [Notice that the complementary submodule N can be taken as a direct sum of some of the original Mi .] (c) implies (a). First we make several observations. (1) If M satisﬁes (c), so does every submodule N . [Let N ≤ M , so that M = N ⊕ N . If V is a submodule of N , hence of M , we have M = V ⊕ W . If x ∈ N , then x = v + w, v ∈ V, w ∈ W , so w = x − v ∈ N (using V ≤ N ). But v also belongs to N , and consequently N = (N ∩ V ) ⊕ (N ∩ W ) = V ⊕ (N ∩ W ).] (2) If D = A ⊕ B ⊕ C, then A = (A + B) ∩ (A + C). [If a + b = a + c, where a, a ∈ A, b ∈ B, c ∈ C, then a − a = b − c, and since D is a direct sum, we have b = c = 0 and a = a . Thus a + b ∈ A.] (3) If N is a nonzero submodule of M , then N contains a simple submodule. [Choose a nonzero x ∈ N . By Zorn’s lemma, there is a maximal submodule V of N such that x∈ / V . By (1) we can write N = V ⊕ V , and V = 0 by choice of x and V . If V is simple, we are ﬁnished, so assume the contrary. Then V contains a nontrivial proper submodule V1 , so by (1) we have V = V1 ⊕ V2 with the Vj nonzero. By (2), V = (V + V1 ) ∩ (V + V2 ). / V + V2 , which contradicts the maximality of V .] Since x ∈ / V , either x ∈ / V + V1 or x ∈ To prove that (c) implies (a), let N be the sum of all simple submodules of M . By (c) we can write M = N ⊕ N . If N = 0, then by (3), N contains a simple submodule V . But

page 2 of Chapter 9

then V ≤ N by deﬁnition of N . Thus V ≤ N ∩ N = 0, a contradiction. Therefore N = 0 and M = N . ♣ 9.1.3 Proposition Nonzero submodules and quotient modules of a semisimple module are semisimple. Proof.The submodule case follows from (1) of the proof of (9.1.2). Let N ≤ M , where M = i Mi with the Mi simple. Applying the canonical map from M to M/N , we have M/N =

(Mi + N )/N.

i

This key idea has come up before; see the proofs of (1.4.4) and (4.2.3). By the second isomorphism theorem, (Mi + N )/N is isomorphic to a quotient of the simple module Mi . But a quotient of Mi is isomorphic to Mi or to zero, and it follows that M/N is a sum of simple modules. By (a) of (9.1.2), M/N is semisimple. ♣ Problems For Section 9.1 1. Regard a ring R as an R-module. Show that R is simple if and only if R is a division ring. 2. Let M be an R-module, with x a nonzero element of M . Deﬁne the R-module homomorphism f : R → Rx by f (r) = rx. Show that the kernel I of f is a proper ideal of R, and R/I is isomorphic to Rx. 3. If M is a nonzero R-module, show that M is simple if and only if M ∼ = R/I for some maximal left ideal I. 4. If M is a nonzero R-module, show that M is simple if and only if M is cyclic (that is, M can be generated by a single element) and every nonzero element of M is a generator. 5. What do simple Z-modules look like? 6. If F is a ﬁeld, what do simple F [X]-modules look like? 7. Let V be an n-dimensional vector space over a ﬁeld k. (Take n ≥ 1 so that V = 0.) If f is an endomorphism (that is, a linear transformation) of V and x ∈ V , deﬁne f x = f (x). This makes V into a module over the endomorphism ring Endk (V ). Show that the module is simple. 8. Show that a nonzero module M is semisimple if and only if every short exact sequence 0 → N → M → P → 0 splits. 9.2 Two Key Theorems If M is a simple R-module, there are strong restrictions on a homomorphism either into or out of M . A homomorphism from one simple R-module to another is very severely restricted, as Schur’s lemma reveals. This very useful result will be important in the proof of Wedderburn’s structure theorem. Another result that will be needed is a theorem of Jacobson that gives some conditions under which a module homomorphism f amounts to multiplication by a ﬁxed element of a ring, at least on part of the domain of f . 9.2.1 Schur’s Lemma (a) If f ∈ HomR (M, N ) where M and N are simple R-modules, then f is either identically 0 or an isomorphism. (b) If M is a simple R-module, then EndR (M ) is a division ring. Proof. (a) The kernel of f is either 0 or M , and the image of f is either 0 or N . If f is not the zero map, then the kernel is 0 and the image is N , so f is an isomorphism. (b) Let f ∈ EndR (M ), f not identically 0. By (a), f is an isomorphism, and therefore is invertible in the endomorphism ring of M . ♣

page 3 of Chapter 9

The next result prepares for Jacobson’s theorem. 9.2.2 Lemma Let M be a semisimple R-module, and let A be the endomorphism ring EndR (M ). [Note that M is an A-module; if g ∈ A we take g • x = g(x), x ∈ M .] If m ∈ M and f ∈ EndA (M ), then there exists r ∈ R such that f (m) = rm. Before proving the lemma, let’s look more carefully at EndA (M ). Suppose that f ∈ EndA (M ) and x ∈ M . If g ∈ A then f (g(x)) = g(f (x)). Thus EndA (M ) consists of those abelian group endomorphisms of M that commute with everything in EndR (M ). In turn, by the requirement that f (rx) = rf (x), EndR (M ) consists of those abelian group endomorphisms of M that commute with R, more precisely with multiplication by r, for each r ∈ R. For this reason, EndA (M ) is sometimes called the double centralizer of R. We also observe that the map taking r ∈ R to multiplication by r is a ring homomorphism of R into EndA (M ). [Again use rf (x) = f (rx).] Jacobson’s theorem will imply that given any f in EndA (M ) and any ﬁnite set S ⊆ M , some g in the image of this ring homomorphism will agree with f on S. Thus in (9.2.2), we can replace the single element m by an arbitrary ﬁnite subset of M . Proof. By (9.1.2) part (c), we can express M as a direct sum Rm ⊕ N . Now if we have a direct sum U = V ⊕ W and u = v + w, v ∈ V, w ∈ W , there is a natural projection of U on V , namely u → v. In the present case, let π be the natural projection of M on Rm. Then π ∈ A and f (m) = f (πm) = πf (m) ∈ Rm. The result follows. ♣ Before proving Jacobson’s theorem, we review some ideas that were introduced in the exercises in Section 4.4. 9.2.3 Comments To specify an R-module homomorphism ψ from a direct sum V ∗ = th component ⊕nj=1 Vj to a direct sum W ∗ = ⊕m i=1 Wi , we must give, for every i and j, the i of the image of vj ∈ Vj . Thus the homomorphism is described by a matrix [ψij ], where th th ψij is a homomorphism from Vj to nWi . The i component of ψ(vj ) is ψij (vj ), so the i component of ψ(v1 + · · · + vn ) is j=1 ψij (vj ). Consequently,  v1 . ψ(v1 + · · · + vn ) = [ψij ]  ..  . vn 

(1)

This gives an abelian group isomorphism between HomR (V ∗ , W ∗ ) and [HomR (Vj , Wi )], the collection of all m by n matrices whose ij entry is an R-module homomorphism from Vj to Wi . If we take m = n and Vi = Wj = V for all i and j, then V ∗ = W ∗ = V n , the direct sum of n copies of V . Then the abelian group isomorphism given by (1) becomes EndR (V n ) ∼ = Mn (EndR (V )),

(2)

the collection of all n by n matrices whose entries areR-endomorphisms of V . Since composition of endomorphisms corresponds to multiplication of matrices, (2) gives a ring isomorphism as well. 9.2.4 Theorem (Jacobson) Let M be a semisimple R-module, and let A be the endomorphism ring EndR (M ). If f ∈ EndA (M ) and m1 , . . . , mn ∈ M , then there exists r ∈ R such that f (mi ) = rmi for all i = 1, . . . , n. Proof. f induces an endomorphism f (n) of M n , the direct sum of n copies of M , via f (n) (m1 + · · · + mn ) = f (m1 ) + · · · + f (mn ) where f (mi ) belongs to the ith copy of M . Thus the matrix that represents f (n) is the scalar matrix f I, where I is an n by n identity matrix. If B = EndR (M n ), then since a

page 4 of Chapter 9

scalar matrix commutes with everything, f (n) ∈ EndB (M n ). If m1 , . . . , mn ∈ M , then by (9.2.2), there exists r ∈ R such that f (n) (m1 + · · · mn ) = r(m1 + · · · mn ). [Note that since M is semisimple, so is M n .] This is equivalent to f (mi ) = rmi for all i. ♣ Before giving a corollary, we must mention that the standard results that every vector space over a ﬁeld has a basis, and any two bases have the same cardinality, carry over if the ﬁeld is replaced by a division ring. Also recall that a module is said to be faithful if its annihilator is 0. 9.2.5 Corollary Let M be a faithful, simple R-module, and let D = EndR (M ), a division ring by (9.2.1(b)). If M is a ﬁnite-dimensional vector space over D, then EndD (M ) ∼ = R, a ring isomorphism. Proof. Let {x1 , . . . , xn } be a basis for M over D. By (9.2.4), if f ∈ EndD (M ), there exists r ∈ R such that f (xi ) = rxi for all i = 1, . . . , n. Since the xi form a basis, we have f (x) = rx for every x ∈ M . Thus the map h from R to EndD (M ) given by r → gr = multiplication by r is surjective. If rx = 0 for all x ∈ M , then since M is faithful, we have r = 0 and h is injective. Since h(rs) = gr ◦ gs = h(r)h(s), h is a ring isomorphism. ♣ Problems For Section 9.2 1. Criticize the following argument. Let M be a simple R-module, and let A=EndR (M ). “Obviously” M is also a simple A-module. For any additive subgroup N of M that is closed under the application of all R-endomorphisms of M is, in particular, closed under multiplication by an element r ∈ R. Thus N is an R-submodule of M , hence is 0 or M . 2. Let M be a nonzero cyclic module. Show that M is simple if and only if ann M , the annihilator of M , is a maximal left ideal. 3. In Problem 2, show that the hypothesis that M is cyclic is essential. 4. Let V = F n be the n-dimensional vector space of all n-tuples with components in the ﬁeld F . If T is a linear transformation on V , then V becomes an F [X]-module via f (X)v = f (T )v. For example, if n = 2, T (a, b) = (0, a), and f (X) = a0 + a1 X + · · · + an X n , then f (X)(1, 0) = a0 (1, 0) + a1 T (1, 0) + a2 T 2 (1, 0) + · · · + an T n (1, 0) = (a0 , 0) + (0, a1 ) = (a0 , a1 ). Show that in this case, V is cyclic but not simple. 5. Suppose that M is a ﬁnite-dimensional vector space over an algebraically closed ﬁeld F , and in addition M is a module over a ring R containing F as a subring. If M is a simple R-module and f is an R-module homomorphism, in particular an F -linear transformation, on M , show that f is multiplication by some ﬁxed scalar λ ∈ F . This result is frequently given as a third part of Schur’s lemma. 6. Let I be a left ideal of the ring R, so that R/I is an R-module but not necessarily a ring. Criticize the following statement: “Obviously”, I annihilates R/I. 9.3 Simple and Semisimple Rings 9.3.1 Deﬁnitions and Comments Since a ring is a module over itself, it is natural to call a ring R semisimple if it is semisimple as an R-module. Our aim is to determine, if possible, how semisimple rings are assembled from simpler components. A plausible idea is that the components are rings that are simple as modules over themselves. But this turns out to be too restrictive, since the components would have to be division rings (Section 9.1, Problem 1). When we refer to a simple left ideal I of R, we will always mean that I is simple as a left R-module. We say that the ring R is simple if R is semisimple and all simple left ideals of R are isomorphic. [The deﬁnition of simple ring varies in the literature. An advantage of

page 5 of Chapter 9

our choice (also favored by Lang and Bourbaki) is that we avoid an awkward situation in which a ring is simple but not semisimple.] Our goal is to show that the building blocks for semisimple rings are rings of matrices over a ﬁeld, or more generally, over a division ring. The next two results give some properties of modules over semisimple rings. 9.3.2 Proposition If R is a semisimple ring, then every nonzero R-module M is semisimple. Proof. By (4.3.6), M is a quotient of a free R-module F . Since F is a direct sum of copies of R (see (4.3.4)), and R is semisimple by hypothesis, it follows from (9.1.2) that F is semisimple. By (9.1.3), M is semisimple. ♣ 9.3.3 Proposition Let I be a simple left ideal in the semisimple ring R, and let M be a simple R-module. Denote by IM the R-submodule of M consisting of all ﬁnite linear combinations i ri xi , ri ∈ I, xi ∈ M . Then either IM = M and I is isomorphic to M , or IM = 0. Proof. If IM = 0, then since M is simple, IM = M . Thus for some x ∈ M we have Ix = 0, and again by simplicity of M , we have Ix = M . Map I onto M by r → rx, and note that the kernel cannot be I because Ix = 0. Since I is simple, the kernel must be 0, so I ∼ = M. ♣ 9.3.4 Beginning the Decomposition Let R be a semisimple ring. We regard two simple left ideals of R as equivalent if they are isomorphic (as R-modules), and we choose a representative Ii , i ∈ T from each equivalence class. We deﬁne the basic building blocks of R as Bi = the sum of all left ideals of R that are isomorphic to Ii . We have a long list of properties of the Bi to establish, and for the sake of economy we will just number the statements and omit the words “Lemma” and “Proof” in each case. We will also omit the end of proof symbol, except at the very end. 9.3.5 If i = j, then Bi Bj = 0. [The product of two left ideals is deﬁned exactly as in (9.3.3).] Apply (9.3.3) with I replaced by Bi and M by Bj . 9.3.6 R = i∈T Bi If r ∈ R, then (r) is a left ideal, which by (9.1.2) and (9.1.3) (or (9.3.2)) is a sum of simple left ideats. 9.3.7 Each Bi is a two-sided ideal. Using (9.3.5) and (9.3.6) we have Bi ⊆ Bi R = Bi

Bj = Bi Bi ⊆ RBi ⊆ Bi .

j

Thus RBi = Bi R = Bi . 9.3.8 R has only ﬁnitely many isomorphism classes of simple left ideals I1 , . . . , It . By (9.3.6), we can write the identity 1 of R as a ﬁnite sum of elements ei ∈ Bi , i ∈ T . t Adjusting the notation if necessary, let 1 = i=1 ei . If r ∈ Bj where j ∈ / {1, . . . , t}, then by (9.3.5), rei = 0 for all i = 1, . . . , t, so r = r1 = 0. Thus Bj = 0 for j ∈ / {1, . . . , t}. t 9.3.9 R = ⊕ti=1 Bi . Thus 1 has a unique representation as i=1 ei , with ei ∈ Bi . By (9.3.6) and (9.3.8), R is the sum of the Bi . If b1 + · · · + bt = 0, with bi ∈ Bi , then 0 = ei (b1 + · · · + bt ) = ei b1 + · · · ei bt = ei bi = (e1 + · · · + et )bi = 1bi = bi .

page 6 of Chapter 9

Therefore the sum is direct. 9.3.10 If bi ∈ Bi , then ei bi = bi = bi ei . Thus ei is the identity on Bi and Bi = Rei = ei R. The ﬁrst assertion follows from the computation in (9.3.9), along with a similar computation with ei multiplying on the right instead of the left. Now Bi ⊆ Rei because bi = bi ei , and Rei ⊆ Bi by (9.3.7) and the fact that ei ∈ Bi . The proof that Bi = ei R is similar. 9.3.11 Each Bi is a simple ring. By the computation in (9.3.7), along with (9.3.10), Bi is a ring (with identity ei ). Let J be a simple left ideal of Bi . By (9.3.5) and (9.3.6), RJ = Bi J = J, so J is a left ideal of R, necessarily simple. Thus J is isomorphic to some Ij , and we must have j = i. [Otherwise, J would appear in the sums deﬁning both Bi and Bj , contradicting (9.3.9).] Therefore Bi has only one isomorphism class of simple left ideals. Now Bi is a sum of simple left ideals of R, and a subset of Bi that is a left ideal of R must be a left ideal of Bi . Consequently, Bi is semisimple and the result follows. 9.3.12 If M is a simple R-module, then M is isomorphic to some Ii . Thus there are only ﬁnitely many isomorphism classes of simple R-modules. In particular, if R is a simple ring, then all simple R-modules are isomorphic. By (9.3.9), R=

t

Bi =

i=1

t

{J : J ∼ = Ii }

i=1

where the J are simple left ideals. Therefore M = RM =

t

Bi M =

i=1

t

{JM : J ∼ = Ii }.

i=1

By (9.3.3), JM = 0 or J ∼ = M . The former cannot hold for all J, since M = 0. Thus M∼ = Ii for some i. If R is a simple ring, then there is only one i, and the result follows. 9.3.13 Let M be a nonzero R-module, so that M is semisimple by (9.3.2). Deﬁne Mi as thesum of all simple submodules of M that are isomorphic to Ii , so that by (9.3.12), t M = i=1 Mi . Then M=

t

Bi M and Bi M = ei M = Mi , i = 1, . . . , t.

i=1

By deﬁnition of Bi , Bi Mj =

{JMj : J ∼ = Ii }

where the J’s are simple left ideals. If N is any simple module involved in the deﬁnition of Mj , then JN is 0 or N , and by (9.3.3), JN = N implies that N ∼ =J ∼ = Ii . But all such N are isomorphic to Ij , and therefore Bi Mj = 0, i = j. Thus Mi = RMi = Bj Mi = Bi Mi j

and Bi M =

Bi Mj = Bi Mi .

j

Consequently, Mi = Bi M = ei RM = ei M (using (9.3.10)), and all that remains is to show that the sum of the Mi is direct. Let x1 + · · · + xt = 0, xi ∈ Mi . Then 0 = ei (x1 + · · · + xt ) = ei xi

page 7 of Chapter 9

since ei xj ∈ Bi Mj = 0 for i = j. Finally, by (9.3.9), ei xi = (e1 + · · · + et )xi = xi . 9.3.14 A semisimple ring R is ring-isomorphic to a direct product of simple rings. This follows from (9.3.9) and (9.3.5). For if ai , bi ∈ Bi , then (a1 + · · · + at )(b1 + · · · + bt ) = a1 b1 + · · · + at bt . ♣ Problems For Section 9.3 In Problems 1 and 2, let M be a semisimple module, so that M is the direct sum of simple modules Mi , i ∈ I. We are going to show that M is a ﬁnite direct sum of simple modules if and only if M is ﬁnitely generated. 1. Suppose that x1 , . . . , xn generate M . It will follow that M is the direct sum of ﬁnitely many of the Mi . How would you determine which Mi ’s are involved? 2. Conversely, assume that M is a ﬁnite sum of simple modules. Show that M is ﬁnitely generated. 3. A left ideal I is said to be minimal if I = 0 and I has no proper subideal except 0. Show that the ring R is semisimple if and only if R is a direct sum of minimal left ideals. 4. Is Z semisimple? 5. Is Zn semisimple? 6. Suppose that R is a ring with the property that every nonzero R-module is semisimple. Show that every R-module M is projective, that is, every exact sequence 0 → A → B → M → 0 splits. Moreover, M is injective, that is, every exact sequence 0 → M → A → B → 0 splits. [Projective and injective modules will be studied in Chapter 10.] 7. For any ring R, show that the following conditions are equivalent. (a) R is semisimple; (b) Every nonzero R-module is semisimple; (c) Every R-module is projective; (d) Every R-module is injective. 9.4 Further Properties of Simple Rings, Matrix Rings, and Endomorphisms To reach the Wedderburn structure theorem, we must look at simple rings in more detail, and supplement what we already know about matrix rings and rings of endomorphisms. 9.4.1 Lemma Let R be any ring, regarded as a left module over itself. If h : R → M is an Rmodule homomorpbism, then for some x ∈ M we have h(r) = rx for every r ∈ R. Moreover, we may choose x = h(1), and the map h → h(1) is an isomorphism of HomR (R, M ) and M . This applies in particular when M = R, in which case h ∈ EndR (R). Proof. The point is that h is determined by what it does to the identity. Thus h(r) = h(r1) = rh(1) so we may take x = h(1). If s ∈ R and h ∈ HomR (R, M ), we take (sh)(r) = h(rs) = rsx. This makes HomR (R, M ) into a left R-module isomorphic to M . (For further discussion of this idea, see the exercises in Section 10.7.) ♣ Notice that although all modules are left R-modules, h is given by multiplication on the right by x. 9.4.2 Corollary Let I and J be simple left ideals of the simple ring R. Then for some x ∈ R we have J = Ix.

page 8 of Chapter 9

Proof. By the deﬁnition of a simple ring (see (9.3.1)), R is semisimple, so by (9.1.2), R = I ⊕ L for some left ideal L. Again by the deﬁnition of a simple ring, I and J are isomorphic (as R-modules). If τ : I → J is an isomorphism and π is the natural projection of R on I, then τ π ∈ EndR (R), so by (9.4.1), there exists x ∈ R such that τ π(r) = rx for every r ∈ R. Allow r to range over I to conclude that J = Ix. ♣ A semisimple ring can be expressed as a direct sum of simple left ideals, by (9.1.2). If the ring is simple, only ﬁnitely many simple left ideals are needed. 9.4.3 Lemma A simple ring R is a ﬁnite direct sum of simple left ideals. Proof. Let R = ⊕j Ij where the Ij are simple left ideals. Changing notation if necessary, we have 1 = y1 + · · · + ym with yj ∈ Ij , j = 1, . . . , m. If x ∈ R, then x = x1 =

m j=1

xyj ∈

m

Ij .

j=1

Therefore R is a ﬁnite sum of the Ij , and the sum is direct because the original decomposition of R is direct. ♣ 9.4.4 Corollary If I is a simple left ideal of the simple ring R, then IR = R. Proof. If J is any simple left ideal of R, then by (9.4.2), J ⊆ IR. By (9.4.3), R is a ﬁnite (direct) sum of simple left ideals, so R ⊆ IR. The reverse inclusion always holds, and the result follows. ♣ We now have some insight into the structure of simple rings. 9.4.5 Proposition If R is a simple ring, then the only two-sided ideals of R are 0 and R. Proof. Let J be a nonzero 2-sided ideal of R. By (9.1.3), J is a semisimple left R-module, so by (9.1.2), J is a sum of simple left ideals of J, hence of R. In particular, J contains a simple left ideal I. Since J is a right ideal, it follows that J = JR. Using (9.4.4), we have J = JR ⊇ IR = R so J = R. ♣ In the literature, a simple ring is often deﬁned as a ring R whose only two-sided ideals are 0 and R, but then extra hypotheses must be added to force R to be semisimple. See the exercises for further discussion. 9.4.6 Corollary Let I be a simple left ideal of the simple ring R, and let M be a simple R-module. Then IM = M and M is faithful. Proof. The ﬁrst assertion follows from a computation that uses associativity of scalar multiplication in a module, along with (9.4.4): M = RM = (IR)M = I(RM ) = IM.

(1)

Now let b belong to the annihilator of M , so that bM = 0. We must show that b = 0. By a computation similar to (1) (using in addition the associativity of ring multiplication), RbRM = RbM = R0 = 0.

(2)

But RbR is a two-sided ideal of R (see (2.2.7)), so by (9.4.5), RbR = 0 or R. In the latter case, M = RM = RbRM = 0 by (2), contradicting the assumption that M is simple. Therefore RbR = 0, in particular, b = 1b1 = 0. ♣

page 9 of Chapter 9

We are now ready to show that a simple ring is isomorphic to a ring of matrices. Let R be a simple ring, and V a simple R-module. [V exists because R is a sum of simple left ideals, and V is unique up to isomorphism by (9.3.12).] Let D = EndR (V ), a division ring by Schur’s lemma (9.2.1(b)). Then (see (9.2.2)), V is a D-module, in other words, a vector space over D. V is a faithful R-module by (9.4.6), and if we can prove that V is ﬁnitedimensional as a vector space over D, then by (9.2.5), R is ring-isomorphic to EndD (V ). If n is the dimension of V over D, then by (4.4.1), EndD (V ) ∼ = Mn (Do ), the ring of n by n o matrices with entries in the opposite ring D . 9.4.7 Theorem Let R be a simple ring, V a simple R-module, and D the endomorphism ring EndR (V ). Then V is a ﬁnite-dimensional vector space over D. If the dimension of this vector space is n, then (by the above discussion), R∼ = EndD (V ) ∼ = Mn (Do ). Proof. Assume that we have inﬁnitely many linearly independent elements x1 , x2 , . . .. Let Im be the left ideal {r ∈ R : rxi = 0 for all i = 1, . . . , m}. Then the Im decrease as m increases, in fact they decrease strictly. [Given any m, let f be a D-linear transformation on V such that f (xi ) = 0 for 1 ≤ i ≤ m and f (xm+1 ) = 0. By Jacobson’s theorem (9.2.4), there exists r ∈ R such that f (xi ) = rxi , i = 1, . . . , m + 1. But then rx1 = · · · = rxm = 0, rxm+1 = 0, so r ∈ Im \ Im+1 .] Write Im = Jm ⊕ Im+1 , as in (9.1.2) part (c). [Recall from (9.1.3) that since R is semisimple, so are all left ideals.] Iterating this process, we construct a left ideal J1 ⊕ J2 ⊕ · · ·, and again by (9.1.2(c)), R = J0 ⊕ J1 ⊕ J2 ⊕ · · · . Therefore 1 is a ﬁnite sum of elements yi ∈ Ji , i = 0, 1, . . . , t. But then R = J0 ⊕ J1 ⊕ · · · ⊕ Jt and it follows that Jt+1 must be 0, a contradiction. ♣ Problems For Section 9.4 Problems 1-5 are the key steps in showing that a ring R is simple if and only if R is Artinian and has no two-sided ideals except 0 and R. Thus if a simple ring is deﬁned as one with no nontrivial two-sided ideals, then the addition of the Artinian condition gives our deﬁnition of simple ring; in particular, it forces the ring to be semisimple. The result that an Artinian ring with no nontrivial two-sided ideals is isomorphic to a matrix ring over a division ring (Theorem 9.4.7) is sometimes called the Wedderburn-Artin theorem. In Problems 1-5, “simple” will always mean simple in our sense. 1. By (9.4.5), a simple ring has no nontrivial two-sided ideals. Show that a simple ring must be Artinian. 2. If R is an Artinian ring, show that there exists a simple R-module. 3. Let R be an Artinian ring with no nontrivial two-sided ideals. Show that R has a faithful, simple R-module. 4. Continuing Problem 3, if V is a faithful, simple R-module, and D = EndR (V ), show that V is a ﬁnite-dimensional vector space over D. 5. Continuing Problem 4, show that R is ring-isomorphic to EndD (V ), and therefore to a matrix ring Mn (Do ) over a division ring. In the next section, we will prove that a matrix ring over a division ring is simple; this concludes the proof that R is simple iﬀ R is Artinian with no nontrivial two-sided ideals. (In the “if” part, semisimplicity of R follows from basic properties of matrix rings; see Section 2.2, Problems 2, 3 and 4.)

page 10 of Chapter 9

6. If an R-module M is a direct sum ⊕ni=1 Mi of ﬁnitely many simple modules, show that M has a composition series. (Equivalently, by (7.5.12), M is Artinian and Noetherian.) 7. Conversely, if M is semisimple and has a composition series, show that M is a ﬁnite direct sum of simple modules. (Equivalently, by Section 9.3, Problems 1 and 2, M is ﬁnitely generated.) 9.5 The Structure of Semisimple Rings We have now done all the work needed for the fundamental theorem. 9.5.1 Wedderburn Structure Theorem Let R be a semisimple ring. (1) R is ring-isomorphic to a direct product of simple rings B1 , . . . , Bt . (2) There are t isomorphism classes of simple R-modules. If V1 , . . . , Vt are representatives of these classes, let Di be the division ring EndR (Vi ). Then Vi is a ﬁnite-dimensional vector space over Di . If ni is the dimension of this vector space, then there is a ring isomorphism Bi ∼ = Mni (Dio ). = EndDi (Vi ) ∼ Consequently, R is isomorphic to the direct product of matrix rings over division rings. Moreover, (3) Bi Vj = 0, i = j; Bi Vi = Vi . Proof. Assertion (1) follows from (9.3.5), (9.3.9) and (9.3.14). By (9.3.8) and (9.3.12), there are t isomorphism classes of simple R-modules. The remaining statements of (2) follow from (9.4.7). The assertions of (3) follow from (9.3.13) and its proof. ♣ Thus a semisimple ring can always be assembled from matrix rings over division rings. We now show that such matrix rings can never combine to produce a ring that is not semisimple. 9.5.2 Theorem The ring Mn (R) of all n by n matrices with entries in the division ring R is simple. Proof. We have done most of the work in the exercises for Section 2.2. Let Ck be the set of matrices whose entries are 0 except perhaps in column k, k = 1 . . . , n. Then Ck is a left ideal of Mn (R), and if any nonzero matrix in Ck belongs to a left ideal I, then Ck ⊆ I. (Section 2.2, Problems 2,3,4.) Thus each Ck is a simple left ideal, and Mn (R), the direct sum of C1 , . . . , Cn , is semisimple. Now let I be a nonzero simple left ideal. A nonzero matrix in I must have a nonzero entry in some column, say column k. Deﬁne f : I → Ck by f (A) = Ak , the matrix obtained from A by replacing every entry except those in column k by 0. Then f is an Mn (R)-module homomorphism, since f (BA) = (BA)k = BAk = Bf (A). By construction, f is not identically 0, so by Schur’s lemma, f is an isomorphism. Since the Ck are mutually isomorphic, all simple left ideals are isomorphic, proving that Mn (R) is simple. ♣ 9.5.3 Informal Introduction to Group Representations A major application of semisimple rings and modules occurs in group representation theory, and we will try to indicate the connection. Let k be any ﬁeld, and let G be a ﬁnite group. We form the group algebra kG, which is a vector space over k with basis vectors corresponding to the elements of G. In general, if G = {x1 , . . . , xm }, the elements of kG are of the form α1 x1 + · · · + αm xm , where the αi belong to k. Multiplication in kG is deﬁned in the natural way; we set (αxi )(βxj ) = αβxi xj

page 11 of Chapter 9

and extend by linearity. Then kG is a ring (with identity 1k 1G ) that is also a vector space over k, and α(xy) = (αx)y = x(αy), α ∈ k, x, y ∈ G, so kG is indeed an algebra over k. [This construction can be carried out with an arbitrary ring R in place of k, and with an arbitrary (not necessarily ﬁnite) group G. The result is the group ring RG, a free R-module with basis G.] Now let V be an n-dimensional vector space over k. We want to describe the situation in which “G acts linearly on V ”. We are familiar with group action (Section 5.1), but we now add the condition that each g ∈ G determines a linear transformation ρ(g) on V . We will write ρ(g)(v) as simply gv or g(v), so that g(αv + βw) = αg(v) + βg(w). Thus we can multiply vectors in V by scalars in G. Since elements of kG are linear combinations of elements of G with coeﬃcients in k, we can multiply vectors in V by scalars in kG. To summarize very compactly, V is a kG-module. Now since G acts on V , (hg)v = h(gv) and 1G v = v, g, h ∈ G, v ∈ V . Thus ρ(hg) = ρ(h)ρ(g), and each ρ(g) is invertible since ρ(g)ρ(g −1 ) = ρ(1G ) = the identity on V . Therefore ρ is a homomorphism from G to GL(V ), the group of invertible linear transformations on V . Multiplication in GL(V ) corresponds to composition of functions. The homomorphism ρ is called a representation of G in V , and n, the dimension of V , is called the degree of the representation. If we like, we can replace GL(V ) by the group of all nonsingular n by n matrices with entries in k. In this case, ρ is called a matrix representation. The above process can be reversed. Given a representation ρ, we can deﬁne a linear action of G on V by gv = ρ(g)(v), and thereby make V a kG-module. Thus representations can be identiﬁed with kG-modules. 9.5.4 The Regular Representation If G has order n, then kG is an n-dimensional vector space over k with basis G. We take V to be kG itself, with gv the product of g and v in kG. As an example, let G = {e, a, a2 }, a cyclic group of order 3. V is a 3-dimensional vector space with basis e, a, a2 , and the action of G on V is determined by ee = e, ea = a, ea2 = a2 ; ae = a, aa = a2 , aa2 = e; a2 e = a2 , a2 a = e, a2 a2 = a. Thus the matrices ρ(g) associated with the elements g ∈ G are      1 0 0 0 0 1 0 2 [e] =  0 1 0  , [a] =  1 0 0  , [a ] =  0 0 0 1 0 1 0 1

1 0 0

 0 1. 0

9.5.5 The Role of Semisimplicity Suppose that ρ is a representation of G in V . Assume that the basis vectors of V can be decomposed into two subsets v(A) and v(B) such that matrix of every g ∈ G has the form

A 0 [g] = . 0 B (The elements of A and B will depend on the particular g, but the dimensions of A and B do not change.) The corresponding statement about V is that V = VA ⊕ V B

page 12 of Chapter 9

where VA and VB are kG-submodules of V . We can study the representation by analyzing its behavior on the simpler spaces VA and VB . Maschke’s theorem, to be proved in the next section, says that under wide conditions on the ﬁeld k, this decomposition process can be continued until we reach subspaces that have no nontrivial kG-submodules. In other words, every kG-module is semisimple. In particular, kG is a semisimple ring, and the Wedderburn structure theorem can be applied to get basic information about representations. We will need some properties of projection operators, and it is convenient to take care of this now. 9.5.6 Deﬁnitions and Comments A linear transformation π on a vector space V [or more generally, a module homomorphism] is called a projection of V (on π(V )) if π is idempotent, that is, π 2 = π. We have already met the natural projection of a direct sum onto a component, but there are other possibilities. For example, let p be the projection of −x+y R2 = R ⊕ R given by p(x, y) = ( x−y 2 , 2 ). Note that π must be the identity on π(V ), since π(π(v)) = π(v). If we choose subspaces carefully, we can regard any projection as natural. 9.5.7 Proposition If π is a projection of V , then V is the direct sum of the image of π and the kernel of π. Proof. Since v = π(v) + (v − π(v)) and π(v − π(v)) = 0, V = im V + ker V . To show that the sum is direct, let v = π(w) ∈ ker π. Then 0 = π(v) = π 2 (w) = π(w) = v, so im π ∩ ker π = 0. ♣ 9.5.8 Example For real numbers x and y, we have (x, y) = (x − cy)(1, 0) + y(c, 1), where c is any ﬁxed real number. Thus R2 = R(1, 0) ⊕ R(c, 1), and if we take p(x, y) = (x − cy, 0), then p is a projection of R2 onto R(1, 0). By varying c we can change the complementary subspace R(c, 1). Thus we have many distinct projections onto the same subspace R(1, 0). Problems For Section 9.5 1. Show that the regular representation is faithful, that is, the homomorphism ρ is injective. 2. Let G be a subgroup of Sn and let V be an n-dimensional vector space over k with basis v(1), . . . , v(n). Deﬁne the action of G on V by g(v(i)) = v(g(i)), i = 1, . . . , n. Show that the action is legal. (V is called a permutation module.) 3. Continuing Problem 2, if n = 4, ﬁnd the matrix of g = (1, 4, 3). 4. Here is an example of how a representation can arise in practice. √ Place an equilateral√triangle in the plane V , with the vertices at v1 = (1, 0), v2 = (− 12 , 12 3) and v3 = (− 12 , − 12 3); note that v1 + v2 + v3 = 0. Let G = D6 be the group of symmetries of the triangle, with g = counterclockwise rotation by 120 degrees and h = reﬂection about the horizontal axis. Each member of D6 is of the form g i hj , i = 0, 1, 2, j = 0, 1, and induces a linear transformation on V . Thus we have a representation of G in V (the underlying ﬁeld k can be taken as R). With v1 and v2 taken as a basis for V , ﬁnd the matrices [g] and [h] associated with g and h. 5. Continue from Problem 4, and switch to the standard basis e1 = v1 = (1, 0), e2 = (0, 1). Changing the basis produces an equivalent matrix representation. The matrix representing the element a ∈ G is now of the form [a] = P −1 [a]P where the similarity matrix P is the same for every a ∈ G (the key point). Find the matrix P corresponding to the switch from {v1 , v2 } to {e1 , e2 }, and the matrices [g] and [h] .

page 13 of Chapter 9

6. Consider the dihedral group D8 , generated by elements R (rotation) and F (reﬂection). We assign to R the 2 by 2 matrix

0 1 A= −1 0 and to F the 2 by 2 matrix

1 B= 0

0 . −1

Show that the above assignment determines a matrix representation of D8 of degree 2. 7. Is the representation of Problem 6 faithful? A very accessible basic text on group representation theory is “Representations and Characters of Groups” by James and Liebeck. 9.6 Maschke’s Theorem We can now prove the fundamental theorem on decomposition of representations. It is useful to isolate the key ideas in preliminary lemmas. 9.6.1 Lemma Let G be a ﬁnite group, and k a ﬁeld whose characteristic does not divide |G| (so that division by |G| is legal). Let V be a kG-module, and ψ a linear transformation on V as a vector space over k. Deﬁne θ : V → V by θ(v) =

1 −1 g ψg(v). |G| g∈G

Then not only is θ a linear transformation on the vector space V , but it is also a kGhomomorphism. Proof. Since ψ is a linear transformation and G acts linearly on V (see (9.5.3)), θ is linear. Now if h ∈ G, then 1 −1 θ(hv) = g ψg(hv). |G| g∈G

As g ranges over all of G, so does gh. Thus we can let x = gh, g −1 = hx−1 , to obtain θ(hv) =

1 −1 hx ψ(xv) = hθ(v) |G| x∈G

and the result follows. ♣ 9.6.2 Lemma In (9.6.1), suppose that ψ is a projection of V on a subspace W that is also a kG-submodule of V . Then θ is also a projection of V on W . Proof. If v ∈ W , then g(v) ∈ W since W is a kG-submodule of V . Thus ψg(v) = g(v) since ψ is a projection on W . By deﬁnition of θ we have θ(v) = v. To prove that θ2 = θ, note that since ψ maps V into the kG-submodule W , it follows from the deﬁnition of θ that θ also maps V into W . But θ is the identity on W , so θ2 (v) = θ(θ(v)) = θ(v) and θ is a projection. Since θ maps into W and is the identity on W , θ is a projection of V on W . ♣ 9.6.3 Maschke’s Theorem Let G be a ﬁnite group, and k a ﬁeld whose characteristic does not divide |G|. If V is a kG-module, then V is semisimple. Proof. Let W be a kG-submodule of V . Ignoring the group algebra for a moment, we can write V = W ⊕ U as vector spaces over k. Let ψ be the natural projection of V on W , and

page 14 of Chapter 9

deﬁne θ as in (9.6.1). By (9.6.1) and (9.6.2), θ is a kG-homomorphism and also a projection of V on W . By (9.5.7), V = im θ ⊕ ker θ = W ⊕ ker θ as kG-modules. By (9.1.2), V is semisimple. ♣ We have been examining the decomposition of a semisimple module into a direct sum of simple modules. Suppose we start with an arbitrary module M , and ask whether M can be expressed as M1 ⊕ M2 , where M1 and M2 are nonzero submodules. If so, we can try to decompose M1 and M2 , and so on. This process will often terminate in a ﬁnite number of steps. 9.6.4 Deﬁnition The module M is decomposable if M = M1 ⊕ M2 , where M1 and M2 are nonzero submodules. Otherwise, M is indecomposable. 9.6.5 Proposition Let M be a module with a composition series; equivalently, by (7.5.12), M is Noetherian and Artinian. Then M can be expressed as a ﬁnite direct sum ⊕ni=1 Mi of indecomposable submodules. Proof. If the decomposition process does not terminate, inﬁnite ascending and descending chains are produced, contradicting the hypothesis. ♣ As the above argument shows, the hypothesis can be weakened to M Noetherian or Artinian. But (9.6.5) is usually stated along with a uniqueness assertion which uses the stronger hypothesis: If M has a composition series and M = ⊕ni=1 Mi = ⊕m j=1 Nj , where the Mi and Nj are indecomposable submodules, then n = m and the Mi are, up to isomorphism, just a rearrangement of the Ni . The full result (existence plus uniqueness) is most often known as the Krull-Schmidt Theorem. [One or more of the names Remak, Azumaya and Wedderburn are sometimes added.] The uniqueness proof is quite long (see, for example, Jacobson’s Basic Algebra II), and we will not need the result. Returning to semisimple rings, there is an asymmetry in the deﬁnition in that a ring is regarded as a left module over itself, so that submodules are left ideals. We can repeat the entire discussion using right ideals, so that we should distinguish between left-semisimple and right-semisimple rings. However, this turns out to be unnecessary. 9.6.6 Theorem A ring R is left-semisimple if and only if it is right-semisimple. Proof. If R is left-semisimple, then by (9.5.1), R is isomorphic to a direct product of matrix rings over division rings. But a matrix ring over a division ring is right-simple by (9.5.2) with left ideals replaced by right ideals. Therefore R is right-semisimple. The reverse implication is symmetrical. ♣ Problems For Section 9.6 1. Let V be the permutation module for G = S3 (see Section 9.5, Problem 2), with basis v1 , v2 , v3 . Give an example of a nontrivial kG-submodule of V . In Problems 2-4, we show that Maschke’s theorem can fail if the characteristic of k divides the order of G. Let G = {1, a, . . . , ap−1 } be a cyclic group of prime order p, and let V be a two-dimensional vector space over the ﬁeld Fp , with basis v1 , v2 . Take the matrix of a as

1 1 [a] = 0 1

so that

1 [a ] = 0 r

and [ap ] is the identity.

r 1

page 15 of Chapter 9

2. Show that W , the one-dimensional subspace spanned by v1 , is a kG-submodule of V . 3. Continuing Problem 2, show that W is the only one-dimensional kG-submodule of V . 4. Continuing Problem 3, show that V is not a semisimple kG-module. 5. Show that a semisimple module is Noetherian iﬀ it is Artinian. 6. Let M be a decomposable R-module, so that M is the direct sum of nonzero submodules M1 and M2 . Show that EndR (M ) contains a nontrivial idempotent e (that is, e2 = e with e not the zero map and not the identity). 7. Continuing from Problem 6, suppose conversely that EndR (M ) contains a nontrivial idempotent e. Show that M is decomposable. (Suggestion: use e to construct idempotents e1 and e2 that are orthogonal, that is, e1 e2 = e2 e1 = 0.) 9.7 The Jacobson Radical There is a very useful device that will allow us to look deeper into the structure of rings. 9.7.1 Deﬁnitions and Comments The Jacobson radical J(R) of a ring R is the intersection of all maximal left ideals of R. More generally, the Jacobson radical J(M ) = JR (M ) of an R-module M is the intersection of all maximal submodules of M . [“Maximal submodule” will always mean “maximal proper submodule”.] If M has no maximal submodule, take J(M ) = M . If M is ﬁnitely generated, then every submodule N of M is contained in a maximal submodule, by Zorn’s lemma. [If the union of a chain of proper submodules is M , then the union contains all the generators, hence some member of the chain contains all the generators, a contradiction.] Taking N = 0, we see that J(M ) is a proper submodule of M . Since R is ﬁnitely generated (by 1R ), J(R) is always a proper left ideal. Semisimplicity of M imposes a severe constraint on J(M ). 9.7.2 Proposition If M is semisimple, then J(M ) = 0. Thus in a sense, the Jacobson radical is an “obstruction” to semisimplicity. Proof. Let N be any simple submodule of M . By (9.1.2), M = N ⊕ N for some submodule N . Now M/N ∼ = N , which is simple, so by the correspondence theorem, N is maximal. Thus J(M ) ⊆ N , and therefore J(M ) ∩ N = 0. Since M is a sum of simple modules (see (9.1.2)), J(M ) = J(M ) ∩ M = 0 ♣ Here is another description of the Jacobson radical. 9.7.3 Proposition J(R) is the intersection of all annihilators of simple R-modules. Proof. By Section 9.1, Problem 3, simple modules are isomorphic to R/I for maximal left ideals I. If r annihilates all simple R-modules, then for every maximal left ideal I, r annihilates R/I, in particular, r annihilates 1 + I. Thus r(1 + I) = I, that is, r ∈ I. Consequently, r ∈ J(R). Conversely, assume r ∈ J(R). If M is a simple R-module, choose any nonzero element x ∈ M . The map fx : R → M given by fx (s) = sx is an epimorphism by simplicity of M . The kernel of fx is the annihilator of x, denoted by ann(x). By the ﬁrst isomorphism theorem, M ∼ = R/ann(x). By simplicity of M , ann(x) is a maximal left ideal, so by hypothesis, r ∈ ∩x∈M ann(x) = ann(M ). Thus r annihilates all simple R-modules. ♣ 9.7.4 Corollary J(R) is a two-sided ideal. Proof. We noted in (4.2.6) that ann(M ) is a two-sided ideal, and the result follows from (9.7.3). ♣ In view of (9.7.4), one might suspect that the Jacobson radical is unchanged if right rather than left ideals are used in the deﬁnition. This turns out to be the case.

page 16 of Chapter 9

9.7.5 Deﬁnitions and Comments The element a ∈ R is left quasi-regular (lqr) if 1 − a has a left inverse, right quasi-regular (rqr) if 1−a has a right inverse, and quasi-regular (qr) if 1−a is invertible. Note that if a is both lqr and rqr, it is qr, because if b(1−a) = (1−a)c = 1, then b = b1 = b(1 − a)c = 1c = c. 9.7.6 Lemma Let I be a left ideal of R. If every element of I is lqr, then every element of I is qr. Proof. If a ∈ I, then we have b(1 − a) = 1 for some b ∈ R. Let c = 1 − b, so that (1 − c)(1 − a) = 1 − a − c + ca = 1. Thus c = ca − a = (c − 1)a ∈ I. By hypothesis, c is lqr, so 1 − c has a left inverse. But we know that (1 − c) has a right inverse (1 − a) [see above], so c is rqr. By (9.7.5), c is qr and 1 − c is the two-sided inverse of 1 − a. ♣ 9.7.7 Proposition The Jacobson radical J(R) is the largest two-sided ideal consisting entirely of quasi-regular elements. Proof. First, we show that each a ∈ J(R) is lqr, so by (9.7.6), each a ∈ J(R) is qr. If 1 − a has no left inverse, then R(1 − a) is a proper left ideal, which is contained in a maximal left ideal I (as in (2.4.2) or (9.7.1)). But then a ∈ I and 1 − a ∈ I, and therefore 1 ∈ I, a contradiction. Now we show that every left ideal (hence every two-sided ideal) I consisting entirely of quasi-regular elements is contained in J(R). If a ∈ I but a ∈ / J(R), then for some maximal left ideal L we have a ∈ / L. By maximality of L, we have I + L = R, so 1 = b + c for some b ∈ I, c ∈ L. But then b is quasi-regular, so c = 1 − b has an inverse, and consequently 1 ∈ L, a contradiction. ♣ 9.7.8 Corollary J(R) is the intersection of all maximal right ideals of R. Proof. We can reproduce the entire discussion beginning with (9.7.1) with left and right ideals interchanged, and reach exactly the same conclusion, namely that the “right”Jacobson radical is the largest two-sided ideal consisting entirely of quasi-regular elements. It follows that the “left”and “right” Jacobson radicals are identical. ♣. We can now use the Jacobson radical to sharpen our understanding of semisimple modules and rings. 9.7.9 Theorem If M is a nonzero R-module, the following conditions are equivalent: (1) M is semisimple and has ﬁnite length, that is, has a composition series; (2) M is Artinian and J(M ) = 0. Proof. (1) implies (2) by (7.5.12) and (9.7.2), so assume M Artinian with J(M ) = 0. The Artinian condition implies that the collection of all ﬁnite intersections of maximal submodules of M has a minimal element N . If S is any maximal submodule of M , then N ∩ S is a ﬁnite intersection of maximal submodules, so by minimality of N , N ∩ S = N , so N ⊆ S. Since J(M ) is the intersection of all such S, the hypothesis that J(M ) = 0 implies that N = 0. Thus for some positive integer n we have maximal submodules M1 , . . . , Mn such that ∩ni=1 Mi = 0. Now M is isomorphic to a submodule of M = ⊕ni=1 (M/Mi ). To see this, map x ∈ M to (x + M1 , . . . , x + Mn ) and use the ﬁrst isomorphism theorem. Since M is a ﬁnite direct sum of simple modules, it is semisimple and has a composition series. (See Section 9.4, Problem 6) By (9.1.3) and (7.5.7), the same is true for M . ♣ 9.7.10 Corollary The ring R is semisimple if and only if R is Artinian and J(R) = 0. Proof. By (9.7.9), it suﬃces to show that if R is semisimple, then it has a composition series. But this follows because R is ﬁnitely generated, hence is a ﬁnite direct sum of simple modules (see Section 9.3, Problem 1). ♣

page 17 of Chapter 9

The Jacobson radical of an Artinian ring has some special properties. 9.7.11 Deﬁnitions and Comments An ideal (or left ideal or right ideal) I of the ring R is nil if each element x ∈ I is nilpotent, that is, xm = 0 for some positive integer m; I is nilpotent if I n = 0 for some positive integer n. Every nilpotent ideal is nil, and the converse holds if R is Artinian, as we will prove. 9.7.12 Lemma If I is a nil left ideal of R, then I ⊆ J(R). Proof. If x ∈ I and xm = 0, then x is quasi-regular; the inverse of 1 − x is 1 + x + x2 + · · · + xm−1 . The result follows from the proof of (9.7.7). ♣ 9.7.13 Proposition If R is Artinian, then J(R) is nilpotent. Thus by (9.7.11) and (9.7.12), J(R) is the largest nilpotent ideal of R, and every nil ideal of R is nilpotent. Proof. Let J = J(R). The sequence J ⊇ J 2 ⊇ · · · stabilizes, so for some n we have J n = J n+1 = · · ·, in particular, J n = J 2n . We claim that J n = 0. If not, then the collection of all left ideals Q of R such that J n Q = 0 is nonempty (it contains J n ), hence has a minimal element N . Choose x ∈ N such that J n x = 0. By minimality of N , J n x = N . Thus there is an element c ∈ J n such that cx = x, that is, (1 − c)x = 0. But c ∈ J n ⊆ J, so by (9.7.7), 1 − c is invertible, and consequently x = 0, a contradiction. ♣ Problems For Section 9.7 1. Show that an R-module is M cyclic if and only if M is isomorphic to R/I for some left ideal I, and in this case we can take I to be ann(M ), the annihilator of M . 2. Show that the Jacobson radical of an R-module M is the intersection of all kernels of homomorphisms from M to simple R-modules. 3. If I = J(R), show that J(R/I) = 0. 4. If f is an R-module homomorphism from M to N , show that f (J(M )) ⊆ J(N ). 5. Assume R commutative, so that J(R) is the intersection of all maximal ideals of R. If a ∈ R, show that a ∈ J(R) if and only if 1 + ab is a unit for every b ∈ R. 6. If N is a submodule of the Jacobson radical of the R-module M , show that J(M )/N = J(M/N ). 9.8 Theorems of Hopkins-Levitzki and Nakayama From Section 7.5, we know that a Noetherian ring need not be Artinian, and an Artinian module need not be Noetherian. But the latter situation can never arise for rings, because of the following result. 9.8.1 Theorem (Hopkins and Levitzki) Let R be an Artinian ring, and M a ﬁnitely generated R-module. Then M is both Artinian and Noetherian. In particular, with M = R, an Artinian ring is Noetherian. Proof. By (7.5.9), M is Artinian. Let J be the Jacobson radical of R. By Section 9.7, Problem 3, the Jacobson radical of R/J is zero, and since R/J is Artinian by (7.5.7), it is semisimple by (9.7.9). Now consider the sequence M0 = M, M1 = JM, M2 = J 2 M, . . . . By (9.7.13), J is nilpotent, so Mn = 0 for some n. Since JMi = Mi+1 , J annihilates Mi /Mi+1 , so by Section 4.2, Problem 6, Mi /Mi+1 is an R/J-module. We claim that each Mi /Mi+1 has a composition series. We can assume that Mi /Mi+1 = 0, otherwise there is nothing to prove. By (9.3.2), Mi /Mi+1 is semisimple, and by (7.5.7), Mi /Mi+1 is Artinian. [Note that submodules of Mi /Mi+1 are the same, whether we use scalars from R or from R/J; see Section 4.2, Problem 6.] By Section 9.6, Problem 5, Mi /Mi+1 is Noetherian, hence has a composition series by (7.5.12).

page 18 of Chapter 9

Now intuitively, we can combine the composition series for the Mi /Mi+1 to produce a composition series for M , proving that M is Noetherian. Formally, Mn−1 ∼ = Mn−1 /Mn has a composition series. Since Mn−2 /Mn−1 has a composition series, so does Mn−2 , by (7.5.7). Iterate this process until we reach M . ♣ We now proceed to a result that has many applications in both commutative and noncommutative algebra. 9.8.2 Nakayama’s Lemma, Version 1 Let M be a ﬁnitely generated R-module, and I a two-sided ideal of R. If I ⊆ J(R) and IM = M , then M = 0. Proof. Assume M = 0, and let x1 , . . . , xn generate M , where n is as small as possible. (Then m n ≥ 1 and the xi are nonzero.) Since xn ∈ M = IM , we can write xn = i=1 bi yi for some n bi ∈ I and yi ∈ M . But yi can be expressed in terms of the generators as yi = j=1 aij xj with aij ∈ R. Thus n xn = bi aij xj = cj xj m

i,j

j=1

where cj = i=1 bi aij . Since I is a right ideal, cj ∈ I ⊆ J(R). (We need I to be a left ideal to make IM a legal submodule of M .) The above equation can be written as (1 − cn )xn =

n−1

cj xj

j=1

and by (9.7.7), 1 − cn is invertible. If n > 1, then xn is a linear combination of the other xi ’s, contradicting the minimality of n. Thus n = 1, in which case (1 − c1 )x1 = 0, so x1 = 0, again a contradiction. ♣ There is another version of Nakayama’s lemma, which we prove after a preliminary result. 9.8.3 Lemma Let N be a submodule of the R-module M , I a left ideal of R. Then M = N + IM if and only if M/N = I(M/N ). Proof. Assume M =N + IM , and let x + N ∈ M/N . Then x = y + z for some y ∈ N and t z ∈ IM . Write z = i=1 ai wi , ai ∈ I, wi ∈ M . It follows that x + N = a1 (w1 + N ) + · · · + at (wt + N ) ∈ I(M/N ). Conversely, assume M/N = I(M/N ), and let x ∈ M . Then x+N =

t

ai (wi + N )

i=1

with ai ∈ I and wi ∈ M . Consequently, x −

t i=1

ai wi ∈ N , so x ∈ N + IM . ♣

9.8.4 Nakayama’s Lemma, Version 2 Let N be a submodule of the R-module M , with M/N ﬁnitely generated over R. [This will be satisﬁed if M is ﬁnitely generated over R.] If I is a two-sided ideal contained in J(R), and M = N + IM , then M = N . Proof. By (9.8.3), I(M/N ) = M/N , so by (9.8.2), M/N = 0, hence M = N . ♣ Here is an application of Nakayama’s lemma. 9.8.5 Proposition Let R be a commutative local ring with maximal ideal J (see (8.5.8)). Let M be a ﬁnitely generated R-module, and let V = M/JM . Then: (i) V is a ﬁnite-dimensional vector space over the residue ﬁeld k = R/J.

page 19 of Chapter 9

(ii) If {x1 + JM, . . . , xn + JM } is a basis for V over k, then {x1 , . . . , xn } is a minimal set of generators for M . (iii) Any two minimal generating sets for M have the same cardinality. Proof. (i) Since J annihilates M/JM , it follows from Section 4.2, Problem 6, that V is a k-module, that is, a vector space over k. Since M is ﬁnitely generated over R, V is ﬁnite-dimensional over k. n (ii) Let N = i=1 Rxi . Since the xi + JM generate V = M/JM , we have M = N + JM . By (9.8.4), M = N , so the xi generate M . If a proper subset of the xi were to generate M , then the corresponding subset of the xi + JM would generate V , contradicting the assumption that V is n-dimensional. (iii) A generating set S for M with more than n elements determines a spanning set for V , which must contain a basis with exactly n elements. By (ii), S cannot be minimal. ♣ Problems For Section 9.8 1. Let a be a nonzero element of the integral domain R. If (at ) = (at+1 ) for some positive integer t, show that a is invertible. 2. Continuing Problem 1, show that every Artinian integral domain is a ﬁeld. 3. If R is a commutative Artinian ring, show that every prime ideal of R is maximal. 4. Let R be a commutative Artinian ring. If S is the collection of all ﬁnite intersections of maximal ideals of R, then S is not empty, hence contains a minimal element I = I1 ∩ I2 ∩ · · · ∩ In , with the Ij maximal. Show that if P is any maximal ideal of R, then P must be one of the Ij . Thus R has only ﬁnitely many maximal ideals. 5. An R-module is projective if it is a direct summand of a free module. We will study projective modules in detail in Section 10.5. We bring up the subject now in Problems 5 and 6 to illustrate a nice application of Nakayama’s lemma. Let R be a commutative local ring, and let M be a ﬁnitely generated projective module over R, with a minimal set of generators {x1 , . . . , xn } (see (9.8.5)). We can assume that for some free module F of rank n, F = M ⊕ N. To justify this, let F be free with basis e1 , . . . , en , and map F onto M via ei → xi , i = 1, . . . , n. If the kernel of the mapping is K, then we have a short exact sequence 0 → K → F → M → 0, which splits since M is projective. [This detail will be covered in (10.5.3).] Let J be the maximal ideal of R, and k = R/J the residue ﬁeld. Show that F/JF ∼ = M/JM ⊕ N/JN. 6. Continue from Problem 5 and show that N/JN = 0. It then follows from Nakayama’s lemma (9.8.2) that N = 0, and therefore M = F . We conclude that a ﬁnitely generated projective module over a commutative local ring is free. 7. We showed in (9.6.6) that there is no distinction between a left and a right-semisimple ring. This is not the case for Noetherian (or Artinian) rings. Let X and Y be noncommuting indeterminates, in other words, XY = Y X, and let Z < X, Y > be the set of all polynomials in X and Y with integer coeﬃcients. [Elements of Z do commute with the indeterminates.] We impose the relations Y 2 = 0 and Y X = 0 to produce the ring R; formally, R = Z < X, Y > /(Y 2 , Y X). Consider I = Z[X]Y , the set of all polynomials f (X)Y, f (X) ∈ Z[X]. Then I is a two-sided ideal of R. Show that if I is viewed as a right ideal, it is not ﬁnitely generated.

page 20 of Chapter 9

Thus R is not right-Noetherian. 8. Viewed as a left R-module, R = Z[X] ⊕ Z[X]Y . Show that R is left-Noetherian. 9. Assume the hypothesis of (9.8.5). If {x1 , . . . , xn } is a minimal generating set for M , show that {x1 , . . . , xn }, where xi = xi + JM , is a basis for M/JM = V . 10. Continuing Problem 9, suppose that {x1 , . . . , xn } and {y1 , . . . , yn } are minimal gener ating sets for M , with yi = j aij xj , aij ∈ R. If A is the matrix of the aij , show that the determinant of A is a unit in R.

page 1 of Chapter 10

CHAPTER 10 INTRODUCING HOMOLOGICAL ALGEBRA Roughly speaking, homological algebra consists of (A) that part of algebra that is fundamental in building the foundations of algebraic topology, and (B) areas that arise naturally in studying (A). 10.1 Categories We have now encountered many algebraic structures and maps between these structures. There are ideas that seem to occur regardless of the particular structure under consideration. Category theory focuses on principles that are common to all algebraic systems. 10.1.1 Deﬁnitions and Comments A category C consists of objects A, B, C, . . . and morphisms f : A → B (where A and B are objects). If f : A → B and g : B → C are morphisms, we have a notion of composition, in other words, there is a morphism gf = g ◦ f : A → C, such that the following axioms are satisﬁed. (i) Associativity: If f : A → B, g : B → C, h : C → D, then (hg)f=h(gf); (ii) Identity: For each object A there is a morphism 1A : A → A such that for each morphism f : A → B, we have f 1A = 1B f = f . A remark for those familiar with set theory: For each pair (A, B) of objects, the collection of morphisms f : A → B is required to be a set rather than a proper class. We have seen many examples: 1. Sets: The objects are sets and the morphisms are functions. 2. Groups: The objects are groups and the morphisms are group homomorphisms. 3. Rings: The objects are rings and the morphisms are ring homomorphisms. 4. Fields: The objects are ﬁelds and the morphisms are ﬁeld homomorphisms [= ﬁeld monomorphisms; see (3.1.2)]. 5. R-mod: The objects are left R-modules and the morphisms are R-module homomorphisms. If we use right R-modules, the corresponding category is called mod-R. 6. Top: The objects are topological spaces and the morphisms are continuous maps. 7. Ab: The objects are abelian groups and the the morphisms are homomorphisms from one abelian group to another. A morphism f : A → B is said to be an isomorphism if there is an inverse morphism g : B → A, that is, gf = 1A and f g = 1B . In Sets, isomorphisms are bijections, and in Top, isomorphisms are homeomorphisms. For the other examples, an isomorphism is a bijective homomorphism, as usual. In the category of sets, a function f is injective iﬀ f (x1 ) = f (x2 ) implies x1 = x2 . But in an abstract category, we don’t have any elements to work with; a morphism f : A → B can be regarded as simply an arrow from A to B. How do we generalize injectivity to an arbitrary category? We must give a deﬁnition that does not depend on elements of a set. Now in Sets, f is injective iﬀ it has a left inverse; equivalently, f is left cancellable, i.e. if f h1 = f h2 , then h1 = h2 . This is exactly what we need, and a similar idea works for surjectivity, since f is surjective iﬀ f is right cancellable, i.e., h1 f = h2 f implies h1 = h2 . 10.1.2 Deﬁnitions and Comments A morphism f is said to be monic if it is left cancellable, epic if it is right cancellable. In all the categories listed in (10.1.1), a morphism f is monic iﬀ f is injective as a mapping of sets. If f is surjective, then it is epic, but the converse can fail. See Problems 2 and 7-10 for some of the details. In the category R-mod, the zero module {0} has the property that for any R-module M , there is a unique module homomorphism from M to {0} and a unique module homomorphism from {0} to M . Here is a generalization of this idea.

page 2 of Chapter 10

10.1.3 Deﬁnitions and Comments Let A be an object in a category. If for every object B, there is a unique morphism from A to B, then A is said to be an initial object. If for every object B there is a unique morphism from B to A, then A is said to be a terminal object. A zero object is both initial and terminal. In the category of sets, there is only one initial object, the empty set. The terminal objects are singletons {x}, and consequently there are no zero objects. In the category of groups, the trivial group consisting of the identity alone is a zero object. We are going to prove that any two initial objects are isomorphic, and similarly for terminal objects. This will be a good illustration of the duality principle, to be discussed next. 10.1.4 Duality If C is a category, the opposite or dual category C op has the same objects as C. The morphisms are those of C with arrows reversed; thus f : A → B is a morphism of C op iﬀ f : B → A is a morphism of C. If the composition gf is permissible in C, then f g is permissible in C op . To see how the duality principle works, let us ﬁrst prove that if A and B are initial objects of C, then A and B are isomorphic. There is a unique morphism f : A → B and a unique morphism g : B → A. But 1A : A → A and 1B : B → B, and it follows that gf = 1A and f g = 1B . The point is that we need not give a separate proof that any two terminal objects are isomorphic. We have just proved the following: If A and B are objects in a category C, and for every object D of C, there is a unique morphism from A to D and there is a unique morphism from B to D, then A and B are isomorphic. Our statement is completely general; it does not involve the properties of any speciﬁc category. If we go through the entire statement and reverse all the arrows, equivalently, if we replace C by C op , we get: If A and B are objects in a category C, and for every object D of C, there is a unique morphism from D to A and there is a unique morphism from D to B, then A and B are isomorphic. In other words, any two terminal objects are isomorphic. If this is unconvincing, just go through the previous proof, reverse all the arrows, and interchange f g and gf . We say that initial and terminal objects are dual. Similarly, monic and epic morphisms are dual. If zero objects exist in a category, then we have zero morphisms as well. If Z is a zero object and A and B arbitrary objects, there is a unique f : A → Z and a unique g : Z → B. The zero morphism from A to B, denoted by 0AB , is deﬁned as gf , and it is independent of the particular zero object chosen (Problem 3). Note that since a zero morphism goes through a zero object, it follows that for an arbitrary morphism h, we have h0 = 0h = 0. 10.1.5 Kernels and Cokernels If f : A → B is an R-module homomorphism, then its kernel is, as we know, {x ∈ A : f (x) = 0}. The cokernel of f is deﬁned as the quotient group B/im(f ). Thus f is injective iﬀ its kernel is 0, and f is surjective iﬀ its cokernel is 0. We will generalize these notions to an arbitrary category that contains zero objects. The following diagram indicates the setup for kernels. C

i →

A

f →

h

g↑

0

B

D We take C to be the kernel of the module homomorphism f , with i the inclusion map. If f g = 0, then the image of g is contained in the kernel of f , so that g actually maps into C. Thus there is a unique module homomorphism h : D → C such that g = ih; simply

page 3 of Chapter 10

take h(x) = g(x) for all x. The key to the generalization is to think of the kernel as the morphism i. This is reasonable because C and i essentially encode the same information. Thus a kernel of the morphism f : A → B is a morphism i : C → A such that: (1) f i = 0. (2) If g : D → A and f g = 0, then there is a unique morphism h : D → C such that g = ih. Thus any map killed by f can be factored through i. If we reverse all the arrows in the above diagram and change labels for convenience, we get an appropriate diagram for cokernels. A

f →

B

0 g↓

p →

C

h

D We take p to be the canonical map of B onto the cokernel of f , so that C = B/im (f ). If gf = 0, then the image of f is contained in the kernel of g, so by the factor theorem, there is a unique homomorphism h such that g = hp. In general, a cokernel of a morphism f : A → B is a morphism p : B → C such that: (1 ) pf = 0. (2 ) If g : B → D and gf = 0, then there is a unique morphism h : C → D such that g = hp. Thus any map that kills f can be factored through p. Since going from kernels to cokernels simply involves reversing arrows, kernels and cokernels are dual. Note, however, that in an arbitrary category with 0, kernels and cokernels need not exist for arbitrary morphisms. But every monic has a kernel and (by duality) every epic has a cokernel; see Problem 5. Problems For Section 10.1 1. Show that in any category, the identity and inverse are unique. 2. In the category of rings, the inclusion map i : Z → Q is not surjective. Show, however, that i is epic. 3. Show that the zero morphism 0AB is independent of the particular zero object chosen in the deﬁnition. 4. Show that a kernel must be monic (and by duality, a cokernel must be epic). [In the deﬁnition of kernel, we can assume that i is monic in (1) of (10.1.5), and drop the uniqueness assumption on h. For i monic forces uniqueness of h, by deﬁnition of monic. Conversely, uniqueness of h forces i to be monic, by Problem 4.] 5. Show that in a category with 0, every monic has a kernel and every epic has a cokernel. 6. Show that if i : C → A and j : D → A are kernels of f : A → B, then C and D are isomorphic. (By duality, a similar statement holds for cokernels.) 7. Let f : A → B be a group homomorphism with kernel K, and assume f not injective, so that K = {1}. Let g be the inclusion map of K into A. Find a homomorphism h such that f g = f h but g = h. 8. It follows from Problem 7 that in the category of groups, f monic is equivalent to f injective as a mapping of sets, and a similar proof works in the category of modules. Why does the argument fail in the category of rings? 9. Continue from Problem 8 and give a proof that does work in the category of rings. 10. Let f : M → N be a module homomorphism with nonzero cokernel, so that f is not surjective. Show that f is not epic; it follows that epic is equivalent to surjective in the category of modules.

page 4 of Chapter 10

10.2 Products and Coproducts We have studied the direct product of groups, rings, and modules. It is natural to try to generalize the idea to an arbitrary category, and a proﬁtable approach is to forget (temporarily) the algebraic structure and just look at the cartesian product A = i Ai of a family of sets Ai , i ∈ I. The key property of a product is that if we are given maps fi from a set S into the factors Ai , we can lift the fi into a single map f : S → i Ai . The commutative diagram below will explain the terminology. Ai fi S

→ f

↑ pi

(1)

A

In the picture, pi is the projection of A onto the ith factor Ai . If fi (x) = ai , i ∈ I, we take f (x) = (ai , i ∈ I). It follows that pi ◦ f = fi for all i; this is what we mean by lifting the fi to f . (Notice that there is only one possible lifting, i.e., f is unique.) If A is the direct product of groups Ai and the fi are group homomorphisms, then f will also be a group homomorphism. Similar statements can be made for rings and modules. We can now give a generalization to an arbitrary category. 10.2.1 Deﬁnition A product of objects Ai in a category C is an object A, along with morphisms pi : A → Ai , with the following universal mapping property. Given any object S of C and morphisms fi : S → Ai , there is a unique morphism f : S → A such that pi f = fi for all i. In a deﬁnition via a universal mapping property, we use a condition involving morphisms, along with a uniqueness statement, to specify an object and morphisms associated with that object. We have already seen this idea in connection with kernels and cokernels in the previous section, and in the construction of the tensor product in Section 8.7. Not every category has products (see Problems 1 and 2), but if they do exist, they are essentially unique. (The technique for proving uniqueness is also essentially unique.) 10.2.2 Proposition If (A, pi , i ∈ I) and (B, qi , i ∈ I) are products of the objects Ai , then A and B are isomorphic. Proof. We use the above diagram (1) with S = B and fi = qi to get a morphism f : B → A such that pi f = qi for all i. We use the diagram with S = A, A replaced by B, pi replaced by qi , and fi = pi , to get a morphism h : A → B such that qi h = pi . Thus pi f h = qi h = pi and qi hf = pi f = qi . But pi 1A = pi and qi 1B = qi and it follows from the uniqueness condition in (10.2.1) that f h = 1A and hf = 1B . Formally, we are using the diagram two more times, once with S = A and fi = pi , and once with S = B, A replaced by B, pi replaced by qi , and fi = qi . Thus A and B are isomorphic. ♣ The discussion of diagram (1) indicates that in the categories of groups, abelian groups, rings, and R-modules, products coincide with direct products. But a category can have products that have no connection with a cartesian product of sets; see Problems 1 and 2. Also, in the category of torsion abelian groups (torsion means that every element has ﬁnite order), products exist but do not coincide with direct products; see Problem 5.

page 5 of Chapter 10

The dual of a product is a coproduct, and to apply duality, all we need to do is reverse all the arrows in (1). The following diagram results. Mj fj N

↓ ij ← f

(2)

M

We have changed the notation because it is now proﬁtable to think about modules. Suppose that M is the direct sum of the submodules Mj , and ij is the inclusion map of Mj into M . If the fj are module homomorphisms out of the factors Mj and into a module N , the fj can be lifted to a single map f . If xj1 ∈ Mj1 , . . . , xjr ∈ Mjr , we take f (xj1 + · · · + xjr) = fj1 (xj1 ) + · · · + fjr (xjr ). Lifting means that f ◦ ij = fj for all j. We can now give the general deﬁnition of coproduct. 10.2.3 Deﬁnition A coproduct of objects Mj in a category C is an object M , along with morphisms ij : Mj → M , with the following universal mapping property. Given any object N of C and morphisms fj : Mj → N , there is a unique morphism f : M → N such that f ij = fj for all j Exactly as in (10.2.2), any two coproducts of a given collection objects are isomorphic. The discussion of diagram (2) shows that in the category of R-modules, the coproduct is the direct sum, which is isomorphic to the direct product if there are only ﬁnitely many factors. In the category of sets, the coproduct is the disjoint union. To explain what this means, suppose we have sets Aj , j ∈ J. We candisjointize the Aj by replacing Aj by Aj = {(x, j) : x ∈ Aj }. The coproduct is A = j∈J Aj , with morphisms ij : Aj → A given by ij (aj ) = (aj , j). If for each j we have fj : Aj → B, we deﬁne f : A → B by f (aj , j) = fj (aj ). The coproduct in the category of groups will be considered in the exercises. Problems For Section 10.2 1. Let S be a preordered set, that is, there is a reﬂexive and transitive relation ≤ on S. Then S can be regarded as a category whose objects are the elements of S. If x ≤ y, there is a unique morphism from x to y, and if x ≤ y, there are no morphisms from x to y. Reﬂexivity implies that there is an identity morphism on x, and transitivity implies that associativity holds. Show that a product of the objects xi , if it exists, must be a greatest lower bound of the xi . The greatest lower bound will be unique (not just essentially unique) if S is a partially ordered set, so that ≤ is antisymmetric. 2. Continuing Problem 1, do products always exist? 3. Continuing Problem 2, what can be said about coproducts? 4. If A is an abelian group, let T (A) be the set of torsion elements of A. Show that T (A) is a subgroup of A. 5. Show that in the category of torsion abelian groups, the product of groups Ai is T ( Ai ), the subgroup of torsion elements of the direct product. 6. Assume that we have a collection of groups Gi , pairwise disjoint except for a common identity 1. The free product of the Gi (notation ∗i Gi ) consists of all words (ﬁnite sequences) a1 · · · an where the aj belong to distinct groups. Multiplication is by concatenation with cancellation. For example, with the subscript j indicating membership in Gj , (a1 a2 a3 a4 )(b4 b2 b6 b1 b3 ) = a1 a2 a3 (a4 b4 )b2 b6 b1 b3

page 6 of Chapter 10

and if b4 = a−1 4 , this becomes a1 a2 a3 b2 b6 b1 b3 . The empty word is the identity, and inverses are calculated in the usual way, as with free groups (Section 5.8). In fact a free group on S is a free product of inﬁnite cyclic groups, one for each element of S. Show that in the category of groups, the coproduct of the Gi is the free product. 7. Suppose that products exist in the category of ﬁnite cyclic groups, and suppose that the cyclic group C with generator a is the product of the cyclic groups C1 and C2 with generators a1 and a2 respectively. Show that the projections p1 and p2 associated with the product of C1 and C2 are surjective. 8. By Problem 7, we may assume without loss of generality that pi (a) = ai , i = 1, 2. Show that for some positive integer n, na1 = a1 and na2 = 0. [Take f1 : C1 → C1 to be the identity map, and let f2 : C1 → C2 be the zero map (using additive notation). Lift f1 and f2 to f : C1 → C.] 9. Exhibit groups C1 and C2 that can have no product in the category of ﬁnite cyclic groups. 10.3 Functors We will introduce this fundamental concept with a concrete example. Let HomR (M, N ) be the set of R-module homomorphisms from M to N . As pointed out at the beginning of Section 4.4, HomR (M, N ) is an abelian group. It will also be an R-module if R is a commutative ring, but not in general. We are going to look at HomR (M, N ) as a function of N , with M ﬁxed. 10.3.1 The Functor HomR (M,

)

We are going to construct a mapping from the category of R-modules to the category of abelian groups. Since a category consists of both objects and morphisms, our map will have two parts: (i) Associate with each R-module N the abelian group HomR (M, N ). (ii) Associate with each R-module homomorphism h : N → P a homomorphism h∗ from the abelian group HomR (M, N ) to the abelian group HomR (M, P ). The following diagram suggests how h∗ should be deﬁned. M

f →

N

h → P

Take h∗ (f ) = hf. Note that if h is the identity on N , then h∗ is the identity on HomR (M, N ). Now suppose we have the following situation: M

f →

N

g →

P

h →

Q

Then (hg)∗ (f ) = (hg)f = h(gf ) = h∗ (g∗ (f )), so that (hg)∗ = h∗ g∗ (= h∗ ◦ g∗ ). To summarize, we have a mapping F called a functor that takes an object A in a category C to an object F (A) in a category D; F also takes a morphism h : A → B in C to a morphism h∗ = F (h) : F (A) → F (B) in D. The key feature of F is the functorial property: F (hg) = F (h)F (g) and F (1A ) = 1F (A) . Thus a functor may be regarded as a homomorphism of categories. 10.3.2 The Functor HomR (

, N)

page 7 of Chapter 10

We now look at HomR (M, N ) as a function of M , with N ﬁxed. Here is an appropriate diagram: g h f K → L → M → N If M is an R-module, we take F (M ) to be the abelian group HomR (M, N ). If h : L → M is an R-module homomorphism, we take h∗ = F (h) to be a homomorphism from the abelian group HomR (M, N ) to the abelian group HomR (L, N ), given by h∗ (f ) = f h. It follows that

(hg)∗ (f ) = f (hg) = (f h)g = g ∗ (f h) = g ∗ (h∗ (f ))

hence

(hg)∗ = g ∗ h∗ ,

and if h is the identity on M , then h∗ is the identity on HomR (M, N ). Thus F does not quite obey the functorial property; we have F (hg) = F (g)F (h) instead of F (hg) = F (h)F (g). However, F is a legal functor on the opposite category of R-mod. In the literature, HomR ( , N ) is frequently referred to as a contravariant functor on the original category R-mod, and HomR (M, ) as a covariant functor on R-mod. If we replace the category of R-modules by an arbitrary category, we can still deﬁne functors (called hom functors) as in (10.3.1) and (10.3.2). But we must replace the category of abelian groups by the category of sets. 10.3.3 The Functors M ⊗R

and

⊗R N

To avoid technical complications, we consider tensor products of modules over a commutative ring R. First we discuss the tensor functor T = M ⊗R The relevant diagram is given below. g f N → P → Q If N is an R-module, we take T (N ) = M ⊗R N . If g : N → P is an R-module homomorphism, we set T (g) = 1M ⊗ g : M ⊗R N → M ⊗R P , where 1M is the identity mapping on M [Recall that (1M ⊗ g)(x ⊗ y) = x ⊗ g(y).] Then T (f g) = 1M ⊗ f g = (1M ⊗ f )(1M ⊗ g) = T (f )T (g) and T (1N ) = 1T (N ) so T is a functor from R-mod to R-mod. The functor S = ⊗R N is deﬁned in a symmetrical way. If M is an R-module, then S(M ) = M ⊗R N , and if f : L → M is an R-module homomorphism, then S(f ) : L ⊗R N → M ⊗R N is given by S(f ) = f ⊗ 1N . 10.3.4 Natural Transformations Again we will introduce this idea with an explicit example. The diagram below summarizes the data. tA F (A) → G(A) Ff ↓ F (B)

↓ Gf → tB

G(B)

(1)

page 8 of Chapter 10

We start with abelian groups A and B and a homomorphism f : A → B. We apply the forgetful functor, also called the underlying functor U. This is a fancy way of saying that we forget the algebraic structure and regard A and B simply as sets, and f as a mapping between sets. Now we apply the free abelian group functor F to produce F (A) = FU(A), the free abelian group with A as basis (and similarly for F (B)). Thus F (A) is the direct sumof copies of Z, one copy for each a ∈ A. The elements of F (A) can be represented as a n(a)x(a), n(a) ∈ Z, where x(a) is the member of the direct sum that is 1 in the ath position and 0 elsewhere. [Similarly, we represent elements of B as b n(b)y(b).] The mapping f determines a homomorphism F f : F (A) → F (B), via n(a)x(a) → n(a)y(f (a)). Now let G be the identity functor, so that G(A) = Gf = f . We deﬁne an A, G(B) = B, abelian group homomorphism t : F (A) → A by t ( n(a)x(a)) = n(a)a, and similarly A A we deﬁne tB ( n(b)y(b)) = n(b)b. (Remember that we began with abelian groups A and B.) The diagram (1) is then commutative, because f tA (x(a)) = f (a) and tB [(F f )(x(a))] = tB (y(f (a)) = f (a). To summarize, we have two functors, F and G from the category C to the category D. (In this case, C = D = the category of abelian groups.) For all objects A, B ∈ C and morphisms f : A → B, we have morphisms tA : F (A) → G(A) and tB : F (B) → G(B) such that the diagram (1) is commutative. We say that t is a natural transformation from F to G. If for every object C ∈ C, tC is an isomorphism (not the case in this example), t is said to be a natural equivalence. The key intuitive point is that the process of going from F (A) to G(A) is “natural” in the sense that as we move from an object A to an object B, the essential features of the process remains the same. Problems For Section 10.3 1. Let F : S → T , where S and T are preordered sets. If we regard S and T as categories, as in Section 10.2, Problem 1, what property must F have in order to be a functor? 2. A group may be regarded as a category with a single object 0, with a morphism for each element g ∈ G. The composition of two morphisms is the morphism associated with the product of the elements. If F : G → H is a function from a group G to a group H, and we regard G and H as categories, what property must F have in order to be a functor? 3. We now look at one of the examples that provided the original motivation for the concept of a natural transformation. We work in the category of vector spaces (over a given ﬁeld) and linear transformations. If V is a vector space, let V ∗ be the dual space, that is, the space of linear maps from V to the ﬁeld of scalars, and let V ∗∗ be the dual of V ∗ . If v ∈ V , let v ∈ V ∗∗ be deﬁned by v(f ) = f (v), f ∈ V ∗ . The mapping from v to v is a linear transformation, and in fact an isomorphism if V is ﬁnite-dimensional. Now suppose that f : V → W and g : W → X are linear transformations. Deﬁne f ∗ : W ∗ → V ∗ by f ∗ (α) = αf, α ∈ W ∗ . Show that (gf )∗ = f ∗ g ∗ . 4. The double dual functor takes a vector space V into its double dual V ∗∗ , and takes a linear transformation f : V → W to f ∗∗ : V ∗∗ → W ∗∗ , where f ∗∗ (v ∗∗ ) = v ∗∗ f ∗ . Show that the double dual functor is indeed a functor. 5. Now consider the following diagram. V

tV →

↓ f ∗∗

f↓ W

V ∗∗

→ tW

W ∗∗

page 9 of Chapter 10

We take tV (v) = v, and similarly for tW . Show that the diagram is commutative, so that t is a natural transformation from the identity functor to the double dual functor. In the ﬁnite-dimensional case, we say that there is a natural isomorphism between a vector space and its double dual. “Natural” means coordinate-free in the sense that it is not necessary to choose a speciﬁc basis. In contrast, the isomorphism of V and its single dual V ∗ is not natural. 6. We say that D is a subcategory of C if the objects of D are also objects of C, and similarly for morphisms (and composition of morphisms). The subcategory D is full if every C-morphism f : A → B, where A and B are objects of D (the key point) is also a D-morphism. Show that the category of groups is a full subcategory of the category of monoids. 7. A functor F : C → D induces a map from C-morphisms to D-morphisms; f : A → B is mapped to F f : F A → F B. If this map is injective for all objects A, B of C, we say that F is faithful. If the map is surjective for all objects A, B of C, we say that F is full. (a) The forgetful functor from groups to sets assigns to each group its underlying set, and to each group homomorphism its associated map of sets. Is the forgetful functor faithful? full? (b) We can form the product C × D of two arbitrary categories; objects in the product are pairs (A, A ) of objects, with A ∈ C and A ∈ D. A morphism from (A, A ) to (B, B ) is a pair (f, g), where f : A → B and g : A → B . The projection functor from C × D to C takes (A, A ) to A and (f, g) to f . Is the projection functor faithful? full? 10.4 Exact Functors 10.4.1 Deﬁnitions and Comments We are going to investigate the behavior of the hom and tensor functors when presented with an exact sequence. We will be working in the categories of modules and abelian groups, but exactness properties can be studied in the more general setting of abelian categories, which we now describe very informally. In any category C, let HomC (A, B) (called a “hom set”) be the set of morphisms in C from A to B. [As remarked in (10.1.1), the formal deﬁnition of a category requires that HomC (A, B) be a set for all objects A and B. The collection of all objects of C is a class but need not be a set.] For C to be an abelian category, the following conditions must be satisﬁed. 1. Each hom set is an abelian group. 2. The distributive laws f (g + h) = f g + f h, (f + g)h = f h + gh hold. 3. C has a zero object. 4. Every ﬁnite set of objects has a product and a coproduct. (The existence of ﬁnite coproducts can be deduced from the existence of ﬁnite products, along with the requirements listed so far.) 5. Every morphism has a kernel and a cokernel. 6. Every monic is the kernel of its cokernel. 7. Every epic is the cokernel of its kernel. 8. Every morphism can be factored as an epic followed by a monic. Exactness of functors can be formalized in an abelian category, but we are going to return to familiar ground by assuming that each category that we encounter is R-mod for some R. When R = Z, we have the category of abelian groups. 10.4.2 Left Exactness of HomR (M, 0

→

A

) Suppose that we have a short exact sequence f →

B

g →

We apply the covariant hom functor F =HomR (M, term on the right. We will show that the sequence 0

→

FA

Ff →

FB

C

→ 0

(1)

) to the sequence, dropping the last Fg →

FC

(2)

page 10 of Chapter 10

is exact. A functor that behaves in this manner is said to be left exact. We must show that the transformed sequence is exact at F A and F B. We do this in three steps. (a) F f is monic. Suppose that (F f )(α) = f α = 0. Since f is monic (by exactness of the sequence (1)), α = 0 and the result follows. (b) im F f ⊆ ker F g. If β ∈ im F f , then β = f α for some α ∈ HomR (M, A). By exactness of (1), im f ⊆ ker g, so gβ = gf α = 0α = 0. Thus β ∈ ker g. (c) ker F g ⊆ im F f . If β ∈ ker F g, then gβ = 0, with β ∈ HomR (M, B). Thus if y ∈ M , then β(y) ∈ ker g = im f , so β(y) = f (x) for some x = α(y) ∈ A. Note that x is unique since f is monic, and α ∈ HomR (M, A). Thus β = f α ∈ im F f . ♣ 10.4.3 Left Exactness of HomR ( , N ) The contravariant hom functor G = HomR ( , N ) is a functor on the opposite category, so before applying it to the sequence (1), we must reverse all the arrows. Thus left-exactness of G means that the sequence 0 →

Gg →

GC

GB

Gf →

GA

(3)

is exact. Again we have three steps. (a) Gg is monic. If (Gg)α = αg = 0, then α = 0 since g is epic. (b) im Gg ⊆ ker Gf . If β ∈ im Gg, then β = αg for some α ∈ HomR (C, N ). Thus (Gf )β = βf = αgf = 0, so β ∈ ker Gf . (c) ker Gf ⊆ im Gg. Let β ∈ HomR (B, N ) with β ∈ ker Gf , that is, βf = 0. If y ∈ C, then since g is epic, we have y = g(x) for some x ∈ B. If g(x1 ) = g(x2 ), then x1 − x2 ∈ ker g = im f , hence x1 − x2 = f (z) for some z ∈ A. Therefore β(x1 ) − β(x2 ) = β(f (z)) = 0, so it makes sense to deﬁne α(y) = β(x). Then α ∈ HomR (C, N ) and αg = β, that is, (Gg)α = β. ♣ 10.4.4 Right Exactness of the Functors M⊗R If we apply the functor H = M ⊗R

0

→

A

and

⊗R N

to the exact sequence f →

B

g →

C

→

0

HC

→

0

[see (1) of (10.4.2)], we will show that the sequence

HA

Hf →

HB

Hg →

(4)

is exact. A similar result holds for ⊗R N . A functor that behaves in this way is said to be right exact. Once again, there are three items to prove. (i) Hg is epic.

page 11 of Chapter 10

An element of M ⊗ C is of the form t = i xi ⊗ yi withxi ∈ M and yi ∈ C. Since g is epic, there exists zi ∈ B such that g(zi ) = yi . Thus (1 ⊗ g)( i xi ⊗ zi ) = i xi ⊗ g(zi ) = t. (ii) im Hf ⊆ ker Hg. This is a brief computation: (1 ⊗ g)(1 ⊗ f ) = 1 ⊗ gf = 1 ⊗ 0 = 0. (iii) ker Hg ⊆ im Hf . By (ii), the kernel of 1 ⊗ g contains L = im (1 ⊗ f ), so by the factor theorem, there is a homomorphism g : (M ⊗R B)/L → M ⊗R C such that g(m⊗b+L) = m⊗g(b), m ∈ M, b ∈ B. Let π be the canonical map of M ⊗R B onto (M ⊗R B)/L. Then gπ(m ⊗ b) = g(m ⊗ b + L) = m ⊗ g(b), so gπ = 1 ⊗ g. If we can show that g is an isomorphism, then ker(1 ⊗ g) = ker(gπ) = ker π = L = im(1 ⊗ f ) and we are ﬁnished. To show that g is an isomorphism, we will display its inverse. First let h be the bilinear map from M × C to (M ⊗R B)/L given by h(m, c) = m ⊗ b + L, where g(b) = c. [Such a b exists because g is epic. If g(b) = g(b ) = c, then b − b ∈ ker g = im f , so b − b = f (a) for some a ∈ A. Then m ⊗ b − m ⊗ b = m ⊗ f (a) = (1 ⊗ f )(m ⊗ a) ∈ L, and h is well-deﬁned.] By the universal mapping property of the tensor product, there is a homomorphism h : M ⊗R C → (M ⊗R B)/L such that h(m ⊗ c) = h(m, c) = m ⊗ b + L, where g(b) = c. But g : (M ⊗R B)/L → M ⊗R C and g(m ⊗ b + L) = m ⊗ g(b) = m ⊗ c. Thus h is the inverse of g. ♣ 10.4.5 Deﬁnition A functor that is both left and right exact is said to be exact. Thus an exact functor is one that maps exact sequences to exact sequences. We have already seen one example, the localization functor (Section 8.5, Problems 4 and 5). If we ask under what conditions the hom and tensor functors become exact, we are led to the study of projective, injective and ﬂat modules, to be considered later in the chapter. Problems For Section 10.4 In Problems 1-3, we consider the exact sequence (1) of (10.4.2) with R = Z, so that we are in the category of abelian groups. Take A = Z, B = Q, the additive group of rational numbers, and C = Q/Z, the additive group of rationals mod 1. Let f be inclusion, and g the canonical map. We apply the functor F =HomR (M, ) with M = Z2 . [We will omit the subscript R when R = Z, and simply refer to Hom(M, ).] 1. Show that Hom(Z2 , Q) = 0. 2. Show that Hom(Z2 , Q/Z) = 0. 3. Show that Hom(Z2 , ) is not right exact. In Problems 4 and 5, we apply the functor G =Hom( , N ) to the above exact sequence, with N = Z. 4. Show that Hom(Q, Z) = 0. 5. Show that Hom( , Z) is not right exact. Finally, in Problem 6 we apply the functor H = M ⊗ with M = Z2 .

to the above exact sequence,

page 12 of Chapter 10

6. Show that Z2 ⊗ (and similarly ⊗ Z2 ) is not left exact. 7. Refer to the sequences (1) and (2) of (10.4.2). If (2) is exact for all possible R-modules M , show that (1) is exact. 8. State an analogous result for the sequence (3) of (10.4.3), and indicate how the result is proved. 10.5 Projective Modules Projective modules are direct summands of free modules, and are therefore images of natural projections. Free modules are projective, and projective modules are sometimes but not always free. There are many equivalent ways to describe projective modules, and we must choose one of them as the deﬁnition. In the diagram below and the deﬁnition to follow, all maps are R-module homomorphisms. The bottom row is exact, that is, g is surjective. P h M

→ g

↓f N

→

0

10.5.1 Deﬁnition The R-module P is projective if given f : P → N , and g : M → N surjective, there exists h : P → M (not necessarily unique) such that the diagram is commutative, that is, f = gh. We sometimes say that we have “lifted” f to h. The deﬁnition may look obscure, but the condition described is a familiar property of free modules. 10.5.2 Proposition Every free module is projective. Proof. Let S be a basis for the free module P . By (4.3.6), f is determined by its behavior on basis elements s ∈ S. Since g is surjective, there exists a ∈ M such that g(a) = f (s). Take h(s) = a and extend by linearity from S to P . Since f = gh on S, the same must be true on all of P . ♣ Here is the list of equivalences. 10.5.3 Theorem The following conditions on the R-module P are equivalent. (1) (2) (3) (4)

P is projective. The functor HomR (P, ) is exact. Every short exact sequence 0 → M → N → P → 0 splits. P is a direct summand of a free module.

Proof. (1) is equivalent to (2). In view of the left exactness of F =HomR (P, ) (see (10.4.2)), (2) says that if g : M → N is surjective, then so is F g : F M → F N . But F g maps h : P → M to gh : P → N , so what we must prove is that for an arbitrary morphism f : P → N , there exists h : P → M such that gh = f . This is precisely the deﬁnition of projectivity of P . (2) implies (3). Let 0 → M → N → P → 0 be a short exact sequence, with g : N → P (necessarily surjective). Since P is projective, we have the following diagram. P h N

→ g

↓1 P

→ 0

page 13 of Chapter 10

Thus there exists h : P → N such that gh = 1P , which means that the exact sequence splits (see (4.7.1)). (3) implies (4). By (4.3.6), P is a quotient of a free module, so there is an exact sequence 0 → M → N → P → 0 with N free. By (3), the sequence splits, so by (4.7.4), P is a direct summand of N . (4) implies (1). Let P be a direct summand of the free module F , and let π be the natural projection of F on P ; see the diagram below. F

π →

P

h↓ M

↓f → g

N

→

0

Given f : P → N , we have f π : F → N , so by (10.5.2) there exists h : F → M such that f π = gh. If h is the restriction of h to P , then f = gh . ♣ 10.5.4 Corollary The direct sum P = ⊕Pj is projective if and only if each Pj is projective. Proof. If P is a direct summand of a free module, so is each Pj , and therefore the Pj are projective by (4) of (10.5.3). Conversely, assume that each Pj is projective. Let f : P → N and g : M → N , with g surjective. If ij is the inclusion map of Pj into P , then f ij : Pj → N can be lifted to hj : Pj → M such that f ij = ghj . By the universal mapping property of direct sum (Section 10.2), there is a morphism h : P → M such that hij = hj for all j. Thus f ij = ghij for every j, and it follows from the uniqueness part of the universal mapping property that f = gh. ♣ If we are searching for projective modules that are not free, the following result tells us where not to look. 10.5.5 Theorem A module M over a principal ideal domain R is projective if and only if it is free. Proof. By (10.5.2), free implies projective. If M is projective, then by (4) of (10.5.3), M is a direct summand of a free module. In particular, M is a submodule of a free module, hence is free by (4.6.2) and the discussion following it. ♣ 10.5.6 Examples 1. A vector space over a ﬁeld k is a free k-module, hence is projective. 2. A ﬁnite abelian group G is not a projective Z-module, because it is not free. [If g ∈ G and n = |G|, then ng = 0, so g can never be part of a basis.] 3. If p and q are distinct primes, then R = Zpq = Zp ⊕ Zq . We claim that Zp and Zq are projective but not free R-modules. (As in Example 2, they are not projective Z-modules.) This follows from (4) of (10.5.3) and the fact that any ring R is a free R-module (with basis {1}). Problems For Section 10.5 In Problems 1-5, we are going to prove the projective basis lemma, which states that an R-module P is projective if and only if there are elements xi ∈ P (i ∈ I) and homomorphisms fi : P → R such that for every x ∈ P , fi (x) = 0 for all but ﬁnitely many i and x=

i

fi (x)xi .

page 14 of Chapter 10

The set of xi ’s is referred to as the projective basis. 1. To prove the “only if” part, let P be a direct summand of the free module F with basis {ei }. Take f to be the inclusion map of P into F , and π the natural projection of F onto P . Show how to deﬁne the fi and xi so that the desired results are obtained. 2. To prove the “if” part, let F be a free module with basis {ei , i ∈ I}, and deﬁne π : F → P by π(ei ) = xi . Deﬁne f : P → F by f (x) = i fi (x)ei . Show that πf is the identity on P . 3. Continuing Problem 2, show that P is projective. 4. Assume that P is ﬁnitely generated by n elements. If Rn is the direct sum of n copies of R, show that P is projective iﬀ P is a direct summand of Rn . 5. Continuing Problem 4, show that if P is projective and generated by n elements, then P has a projective basis with n elements. 6. Show that a module P is projective iﬀ P is a direct summand of every module of which it is a quotient. In other words, if P ∼ = M/N , then P is isomorphic to a direct summand of M. 7. In the deﬁnition (10.5.1) of a projective module, give an explicit example to show that the mapping h need not be unique. 10.6 Injective Modules If we reverse all arrows in the mapping diagram that deﬁnes a projective module, we obtain the dual notion, an injective module. In the diagram below, the top row is exact, that is, f is injective. f 0 → N → M g↓

h

E 10.6.1 Deﬁnition The R-module E is injective if given g : N → E, and f : N → M injective, there exists h : M → E (not necessarily unique) such that g = hf . We sometimes say that we have “lifted” g to h. As with projectives, there are several equivalent ways to characterize an injective module. 10.6.2 Theorem The following conditions on the R-module E are equivalent. (1) E is injective. (2) The functor HomR ( , E) is exact. (3) Every exact sequence 0 → E → M → N → 0 splits. Proof. (1) is equivalent to (2). Refer to (3) of (10.4.3), (1) of (10.4.2) and the deﬁnition of the contravariant hom functor in (10.3.2) to see what (2) says. We are to show that if f : N → M is injective, then f ∗ : HomR (M, E) → HomR (N, E) is surjective. But f ∗ (h) = hf , so given g : N → E, we must produce h : M → E such that g = hf . This is precisely the deﬁnition of injectivity. (2) implies (3). Let 0 → E → M → N → 0 be a short exact sequence, with f : E → M (necessarily injective). Since E is an injective module, we have the following diagram: 0

→

E

f →

1↓

g

E

M

page 15 of Chapter 10

Thus there exists g : M → E such that gf = 1E , which means that the exact sequence splits. (3) implies (1). Given g : N → E, and f : N → M injective, we form the pushout of f and g, which is a commutative square as indicated in the diagram below. f →

N

↓ g

g↓ E

M

→ f

Q

Detailed properties of pushouts are developed in the exercises. For the present proof, all we need to know is that since f is injective, so is f . Thus the sequence

0

→ E

f →

Q →

Q/im f

→

0

is exact. By (3), there exists h : Q → E such that hf = 1E . We now have hg : M → E with hg f = hf g = 1E g = g, proving that E is injective. ♣ We proved in (10.5.4) that a direct sum of modules is projective iﬀ each component is projective. The dual result holds for injectives. 10.6.3 Proposition A direct product j Ej of modules is injective iﬀ each Ej is injective. Consequently, a ﬁnite direct sum is injective iﬀ each summand is injective. Proof. If f : N → M is injective and g : N → i Ei , let gi = pi g, where pi is the projection of the direct product on Ei . Then ﬁnding h : M → i Ei such that g = hf is equivalent to ﬁnding, for each i, a morphism hi : M → Ei such that gi = hi f . [If pi g = hi f = pi hf for every i, then g = hf by the uniqueness part of the universal mapping property for products.] The last assertion holds because the direct sum of ﬁnitely many modules coincides with the direct product. ♣ In checking whether an R-module E is injective, we are given g : N → E, and f : N → M , with f injective, and we must lift g to h : M → E with g = hf . The next result drastically reduces the collection of maps f and g that must be examined. We may take M = R and restrict N to a left ideal I of R, with f the inclusion map. 10.6.4 Baer’s Criterion The R-module E is injective if and only if every R-homomorphism f : I → E, where I is a left ideal of R, can be extended to an R-homomorphism h : R → E. Proof. The “only if” part follows from the above discussion, so assume that we are given g : N → E and f : N → M , where (without loss of generality) f is an inclusion map. We must extend g to h : M → E. A standard Zorn’s lemma argument yields a maximal extension g0 in the sense that the domain M0 of g0 cannot be enlarged. [The partial ordering is (g1 , D1 ) ≤ (g2 , D2 ) iﬀ D1 ⊆ D2 and g1 = g2 on D1 .] If M0 = M , we are ﬁnished, so assume x ∈ M \ M0 . Let I be the left ideal {r ∈ R : rx ∈ M0 }, and deﬁne h0 : I → E by h0 (r) = g0 (rx). By hypothesis, h0 can be extended to h0 : R → E. Let M1 = M0 + Rx and deﬁne h1 : M1 → E by h1 (x0 + rx) = g0 (x0 ) + rh0 (1). To show that h1 is well deﬁned, assume x0 + rx = y0 + sx, with x0 , y0 ∈ M0 and r, s ∈ R. Then (r − s)x = y0 − x0 ∈ M0 , so r − s ∈ I. Using the fact that h0 extends h0 , we have g0 (y0 − x0 ) = g0 ((r − s)x) = h0 (r − s) = h0 (r − s) = (r − s)h0 (1)

page 16 of Chapter 10

and consequently, g0 (x0 ) + rh0 (1) = g0 (y0 ) + sh0 (1) and h1 is well deﬁned. If x0 ∈ M0 , take r = 0 to get h1 (x0 ) = g0 (x0 ), so h1 is an extension of g0 to M1 ⊃ M0 , contradicting the maximality of g0 . We conclude that M0 = M . ♣ Since free modules are projective, we can immediately produce many examples of projective modules. The primary source of injective modules lies a bit below the surface. 10.6.5 Deﬁnitions and Comments Let R be an integral domain. The R-module M is divisible if each y ∈ M can be divided by any nonzero element r ∈ R, that is, there exists x ∈ M such that rx = y. For example, the additive group of rational numbers is a divisible abelian group, as is Q/Z, the rationals mod 1. The quotient ﬁeld of any integral domain (regarded as an abelian group) is divisible. A cyclic group of ﬁnite order n > 1 can never be divisible, since it is not possible to divide by n. The group of integers Z is not divisible since the only possible divisors of an arbitrary integer are ±1. It follows that a nontrivial ﬁnitely generated abelian group, a direct sum of cyclic groups by (4.6.3), is not divisible. It follows from the deﬁnition that a homomorphic image of a divisible module is divisible, hence a quotient or a direct summand of a divisible module is divisible. Also, a direct sum of modules is divisible iﬀ each component is divisible. 10.6.6 Proposition If R is any integral domain, then an injective R-module is divisible. If R is a PID, then an R-module is injective if and only if it is divisible. Proof. Assume E is injective, and let y ∈ E, r ∈ R, r = 0. Let I be the ideal Rr, and deﬁne an R-homomorphism f : I → E by f (tr) = ty. If tr = 0, then since R is an integral domain, t = 0 and f is well deﬁned. By (10.6.4), f has an extension to an R-homomorphism h : R → E. Thus y = f (r) = h(r) = h(r1) = rh(1) so division by r is possible and E is divisible. Conversely, assume that R is a PID and E is divisible. Let f : I → E, where I is an ideal of R. Since R is a PID, I = Rr for some r ∈ R. We have no trouble extending the zero mapping, so assume r = 0. Since E is divisible, there exists x ∈ E such that rx = f (r). Deﬁne h : R → E by h(t) = tx. If t ∈ R, then h(tr) = trx = tf (r) = f (tr) so h extends f , proving E injective. ♣ Problems For Section 10.6 We now describe the construction of the pushout of two module homomorphisms f : A → C and g : A → B; refer to Figure 10.6.1. Take D = (B ⊕ C)/W, where W = {(g(a), −f (a)) : a ∈ A}, and

g (c) = (0, c) + W,

f (b) = (b, 0) + W.

In Problems 1-6, we study the properties of this construction.

A

f C

g

g' g'' B

f'

D h

f'' Figure 10.6.1

E

page 17 of Chapter 10

1. Show that the pushout square ACDB is commutative, that is, f g = g f . 2. Suppose we have another commutative pushout square ACEB with maps f : B → E and g : C → E, as indicated in Figure 10.6.1. Deﬁne h : D → E by h((b, c) + W ) = g (c) + f (b). Show that h is well deﬁned. 3. Show that h makes the diagram commutative, that is, hg = g and hf = f . 4. Show that if h : D → E makes the diagram commutative, then h = h. The requirements stated in Problems 1,3 and 4 can be used to deﬁne the pushout via a universal mapping property. The technique of (10.2.2) shows that the pushout object D is unique up to isomorphism. 5. If f is injective, show that f is also injective. By symmetry, the same is true for g and g . 6. If f is surjective, show that f is surjective. By symmetry, the same is true for g and g . Problems 7-10 refer to the dual construction, the pullback, deﬁned as follows (see Figure 10.6.2). Given f : A → B and g : C → B, take D = {(a, c) ∈ A ⊕ C : f (a) = g(c)} and

g (a, c) = a, f (a, c) = c.

E g'' h A

D g'

f'' f'

C

f

g

B

Figure 10.6.2 7. Show that the pullback square DABC is commutative, that is, f g = gf . 8. If we have another commutative pullback square EABC with maps f : E → C and g : E → A, show that there is a unique h : E → D that makes the diagram commutative, that is, g h = g and f h = f . 9. If f is injective, show that f is injective. By symmetry, the same is true for g and g . 10. If f is surjective, show that f is surjective. By symmetry, the same is true for g and g . 11. Let R be an integral domain with quotient ﬁeld Q, and let f be an R-homomorphism from an ideal I of R to Q. Show that f (x)/x is constant for all nonzero x ∈ I. 12. Continuing Problem 11, show that Q is an injective R-module. 10.7 Embedding into an Injective Module

page 18 of Chapter 10

We know that every module is a quotient of a projective (in fact free) module. In this section we prove the more diﬃcult dual statement that every module can be embedded in an injective module. (To see that quotients and submodules are dual, reverse all the arrows in a short exact sequence.) First, we consider abelian groups. 10.7.1 Proposition Every abelian group can be embedded in a divisible abelian group. Proof. If A is an abelian group, then A is a quotient of a free abelian group F , say A ∼ = F/B. Now F is a direct sum of copies of Z, hence F can be embedded in a direct sum D of copies of Q, the additive group of rationals. It follows that F/B can be embedded in D/B; the embedding is just the inclusion map. By (10.6.5), D/B is divisible, and the result follows. ♣ 10.7.2 Comments In (10.7.1), we used Q as a standard divisible abelian group. It would be very desirable to have a canonical injective R-module. First, we consider H = HomZ (R, A), the set of all abelian group homomorphisms from the additive group of the ring R to the abelian group A. If we are careful, we can make this set into a left R-module. The abelian group structure of H presents no diﬃculties, but we must also deﬁne scalar multiplication. If f ∈ H and s ∈ R, we set (sf )(r) = f (rs), r ∈ R. Checking the module properties is routine except for associativity: ((ts)f )(r) = f (rts), and (t(sf ))(r) = (sf )(rt) = f (rts) so (ts)f = t(sf ). Notice that sf is an abelian group homomorphism, not an R-module homomorphism. Now if A is an R-module and B an abelian group, we claim that

equivalently,

HomZ (A, B) ∼ = HomR (A, HomZ (R, B)),

(1)

HomZ (R ⊗R A, B) ∼ = HomR (A, HomZ (R, B)).

(2)

This is a special case of adjoint associativity: If S MR ,

R N,

SP ,

then

HomS (M ⊗R N, P ) ∼ = HomR (N, HomS (M, P )). Thus if F is the functor M ⊗R

and G is the functor HomS (M,

HomS (F N, P ) ∼ = HomR (N, GP )

(3)

), then (4)

which is reminiscent of the adjoint of a linear operator on an inner product space. We say that F and G are adjoint functors, with F left adjoint to G and G right adjoint to F . [There is a technical naturality condition that is added to the deﬁnition of adjoint functors, but we will not pursue this since the only adjoints we will consider are hom and tensor.] Before giving the formal proof of (3), we will argue intuitively. The left side of the equation describes all biadditive, R-balanced maps from M ×N to P . If (x, y) → f (x, y), x ∈ M, y ∈ N , is such a map, then f determines a map g from N to HomS (M, P ), namely, g(y)(x) = f (x, y). This is a variation of the familiar fact that a bilinear map amounts to a family of linear maps. Now if s ∈ S, then g(y)(sx) = f (sx, y), which need not equal sf (x, y), but if we factor f through the tensor product M ⊗R N , we can then pull out the s. Thus g(y) ∈ HomS (M, P ). Moreover, g is an R-module homomorphism, because

page 19 of Chapter 10

g(ry)(x) = f (x, ry), and we can factor out the r by the same reasoning as above. Since g determines f , the correspondence between f and g is an isomorphism of abelian groups. To prove (3), let f : M ⊗R N → P be an S-homomorphism. If y ∈ N , deﬁne fy : M → P by fy (x) = f (x ⊗ y), and deﬁne ψ(f ) : N → HomS (M, P ) by y → fy . [HomS (M, P ) is a left R-module by Problem 1.] (a) ψ(f ) is an R-homomorphism: ψ(f )(y1 + y2 ) = fy1 +y2 = fy1 + fy2 = ψ(f )(y1 ) + ψ(f )(y2 ); ψ(f )(ry) = fry = rfy = (rψ(f ))(y). [fry (x) = f (x ⊗ ry) and (rfy )(x) = fy (xr) = f (xr ⊗ y).] (b) ψ is an abelian group homomorphism: We have fy (x) + gy (x) = f (x ⊗ y) + g(x ⊗ y) = (f + g)(x ⊗ y) = (f + g)y (x) so ψ(f + g) = ψ(f ) + ψ(g). (c) ψ is injective: If ψ(f ) = 0, then fy = 0 for all y ∈ N , so f (x ⊗ y) = 0 for all x ∈ M and y ∈ N . Thus f is the zero map. (d) If g ∈ HomR (N, HomS (M, P )), deﬁne ϕg : M × N → P by ϕg (x, y) = g(y)(x). Then ϕg is biadditive and R-balanced: By deﬁnition, ϕg is additive in each coordinate, and [see Problem 1] ϕg (xr, y) = g(y)(xr) = (rg(y))(x) = g(ry)(x) = ϕg (x, ry). (e) ψ is surjective: By (d), there is a unique S-homomorphism β(g) : M ⊗R N → P such that β(g)(x⊗y) = ϕg (x, y) = g(y)(x), x ∈ M, y ∈ N . It follows that ψ(β(g)) = g, because ψ(β(g))(y) = β(g)y , where β(g)y (x) = β(g)(x ⊗ y) = g(y)(x). Thus β(g)y = g(y) for all y in N , so ψβ is the identity and ψ is surjective. It follows that ψ is an abelian group isomorphism. This completes the proof of (3). NR ,

Another adjointness result, which can be justiﬁed by similar reasoning, is that if R MS , PS , then HomS (N ⊗R M, P ) ∼ = HomR (N, HomS (M, P ))

which says that F =

⊗R M and G =HomS (M,

(5)

) are adjoint functors.

10.7.3 Proposition If E is a divisible abelian group, then HomZ (R, E) is an injective left R-module. Proof. By (10.6.2), we must prove that HomR ( , HomZ (R, E)) is exact. As in the proof of (1) implies (2) in (10.6.2), if 0 → N → M is exact, we must show that HomR (M, HomZ (R, E)) → HomR (N, HomZ (R, E)) → 0 is exact. By (1) of (10.7.2), this is equivalent to showing that HomZ (M, E) → HomZ (N, E) → 0 is exact. [As indicated in the informal discussion in (10.7.2), this replacement is allowable because a bilinear map can be regarded as a family of linear maps. A formal proof would invoke the naturality condition referred to in (10.7.2).] Since E is an injective Z-module, the result now follows from (10.6.2). ♣

page 20 of Chapter 10

We can now prove the main result. 10.7.4 Theorem If M is an arbitrary left R-module, then M can be embedded in an injective left R-module. Proof. If we regard M as an abelian group, then by (10.7.1), we can assume that M is a subset of the divisible abelian group E. We will embed M in the injective left R-module N = HomZ (R, E) (see (10.7.3)). If m ∈ M , deﬁne f (m) : R → E by f (m)(r) = rm. Then f : M → N , and we claim that f is an injective R-module homomorphism. If f (m1 ) = f (m2 ), then rm1 = rm2 for every r ∈ R, and we take r = 1 to conclude that m1 = m2 , proving injectivity. To check that f is an R-homomorphism, note that if r, s ∈ R and m ∈ M , then f (sm)(r) = rsm and (sf (m))(r) = f (m)(rs) = rsm by deﬁnition of scalar multiplication in the R-module N ; see (10.7.2). ♣ It can be shown that every module M has an injective hull, that is, there is a smallest injective module containing M . Problems For Section 10.7 1. If R MS and R N , show that HomR (M, N ) is a left S-module via (sf )(m) = f (ms). 2. If R MS and NS , show that HomS (M, N ) is a right R-module via (f r)(m) = f (rm). 3. If MR and S NR , show that HomR (M, N ) is a left S-module via (sf )(m) = sf (m). 4. If S M and S NR , show that HomS (M, N ) is a right R-module via (f r)(m) = f (m)r. 5. A useful mnemonic device for remembering the result of Problem 1 is that since M and N are left R-modules, we write the function f on the right of its argument. The result is m(sf ) = (ms)f , a form of associativity. Give similar devices for Problems 2,3 and 4. Note also that in Problem 1, M is a right S-module, but HomR (M, N ) is a left Smodule. The reversal might be expected, because the hom functor is contravariant in its ﬁrst argument. A similar situation occurs in Problem 2, but in Problems 3 and 4 there is no reversal. Again, this might be anticipated because the hom functor is covariant in its second argument. 6. Let R be an integral domain with quotient ﬁeld Q. If M is a vector space over Q, show that M is a divisible R-module. 7. Conversely, if M is a torsion-free divisible R-module, show that M is a vector space over Q. 8. If R is an integral domain that is not a ﬁeld, and Q is the quotient ﬁeld of R, show that HomR (Q, R) = 0. 10.8 Flat Modules 10.8.1 Deﬁnitions and Comments We have seen that an R-module M is projective iﬀ its covariant hom functor is exact, and M is injective iﬀ its contravariant hom functor is

page 21 of Chapter 10

exact. It is natural to investigate the exactness of the tensor functor M ⊗R , and as before we avoid complications by assuming all rings commutative. We say that M is ﬂat if M ⊗R is exact. Since the tensor functor is right exact by (10.4.4), an equivalent statement is that if f : A → B is an injective R-module homomorphism, then 1⊗f :M ⊗A→M ⊗B is injective. In fact it suﬃces to consider only R-modules A and B that are ﬁnitely generated. This can be deduced from properties of direct limits to be considered in the next section. [Any module is the direct limit of its ﬁnitely generated submodules (10.9.3, Example 2). The tensor product commutes with direct limits (Section 10.9, Problem 2). The direct limit is an exact functor (Section 10.9, Problem 4).] A proof that does not involve direct limits can also be given; see Rotman, “An Introduction to Homological Algebra”, page 86. 10.8.2 Example Since Z2 ⊗Z Z-module.

is not exact (Section 10.4, Problem 6), Z2 is not a ﬂat

The next result is the analog for ﬂat modules of property (10.5.4) of projective modules. 10.8.3 Proposition The direct sum ⊕i Mi is ﬂat if and only if each Mi is ﬂat. Proof. Let f : A → B be an injective R-homomorphism. In view of (8.8.6(b)), investigating the ﬂatness of the direct sum amounts to analyzing the injectivity of the mapping g : ⊕i (Mi ⊗ A) → ⊕i (Mi ⊗ B) given by xi1 ⊗ a1 + · · · + xin ⊗ an → xi1 ⊗ f (a1 ) + · · · + xin ⊗ f (an ). The map g will be injective if and only if all component maps xi ⊗ ai → xi ⊗ f (ai ) are injective. This says that the direct sum is ﬂat iﬀ each component is ﬂat. ♣ We now examine the relation between projectivity and ﬂatness. 10.8.4 Proposition R is a ﬂat R-module. Proof. If f : A → B is injective, we must show that (1 ⊗ f ) : R ⊗R A → R ⊗R B is injective. But by (8.7.6), R ⊗R M ∼ = M via r ⊗ x → rx. Thus the following diagram is commutative. R ⊗R A

1⊗f →

|| A

R ⊗R B ||

→ f

B

Therefore injectivity of 1 ⊗ f is equivalent to injectivity of f , and the result follows. ♣ 10.8.5 Corollary Every projective module, hence every free module, is ﬂat. Proof. By (10.8.3) and (10.8.4), every free module is ﬂat. Since a projective module is a direct summand of a free module, it is ﬂat by (10.8.3). ♣ Flat abelian groups can be characterized precisely. 10.8.6 Theorem A Z-module is ﬂat iﬀ it is torsion-free. Proof. Suppose that M is a Z-module that is not torsion-free. Let x ∈ M be a nonzero element such that nx = 0 for some positive integer n. If f : Z → Z is multiplication by n, then (1 ⊗ f ) : M ⊗ Z → M ⊗ Z is given by y ⊗ z → y ⊗ nz = ny ⊗ z

page 22 of Chapter 10

so that (1 ⊗ f )(x ⊗ 1) = nx ⊗ 1 = 0. Since x ⊗ 1 corresponds to x under the isomorphism between M ⊗ Z and M , x ⊗ 1 = 0, and 1 ⊗ f is not injective. Therefore M is not ﬂat. The discussion in (10.8.1) shows that in checking ﬂatness of M , we can restrict to ﬁnitely generated n submodules of M . [We are examining equations of the form (1 ⊗ f )(t) = 0, where t = i=1 xi ⊗ yi , xi ∈ M, yi ∈ A.]Thus without loss of generality, we can assume that M is a ﬁnitely generated abelian group. If M is torsion-free, then by (4.6.5), M is free and therefore ﬂat by (10.8.5). ♣ 10.8.7 Corollary The additive group of rationals Q is a ﬂat but not projective Z-module. Proof. Since Q is torsion-free, it is ﬂat by (10.8.6). If Q were projective, it would be free by (10.5.5). This is a contradiction (see Section 4.1, Problem 5). ♣ Sometimes it is desirable to change the underlying ring of a module; the term base change is used in these situations. 10.8.8 Deﬁnitions and Comments If f : R → S is a ring homomorphism and M is an S-module, we can create an R-module structure on M by rx = f (r)x, r ∈ R, x ∈ M . This is a base change by restriction of scalars. If f : R → S is a ring homomorphism and M is an R-module, we can make S ⊗R M into an S-module via s(s ⊗ x) = ss ⊗ x, s, s ∈ S, x ∈ M. This is a base change by extension of scalars. Note that S is an R-module by restriction of scalars, so the tensor product makes sense. What we are doing is allowing linear combinations of elements of M with coeﬃcients in S. This operation is very common in algebraic topology. In the exercises, we will look at the relation between base change and ﬂatness. There will also be some problems on ﬁnitely generated algebras, so let’s deﬁne these now. 10.8.9 Deﬁnition The R-algebra A is ﬁnitely generated if there are elements a1 , . . . , an ∈ A such that every element of A is a polynomial in the ai . Equivalently, the algebra homomorphism from the polynomial ring R[X1 , . . . , Xn ] → A determined by Xi → ai , i = 1, . . . , n, is surjective. Thus A is a quotient of the polynomial ring. It is important to note that if A is ﬁnitely generated as an R-module, then it is ﬁnitely generated as an R-algebra. [If a = r1 a1 + · · · + rn an , then a is certainly a polynomial in the ai .] Problems For Section 10.8 1. Give an example of a ﬁnitely generated R-algebra that is not ﬁnitely generated as an R-module. 2. Show that R[X] ⊗R R[Y ] ∼ = R[X, Y ]. 3. Show that if A and B are ﬁnitely generated R-algebras, so is A ⊗R B. 4. Let f : R → S be a ring homomorphism, and let M be an S-module, so that M is an R-module by restriction of scalars. If S is a ﬂat R-module and M is a ﬂat S-module, show that M is a ﬂat R-module. 5. Let f : R → S be a ring homomorphism, and let M be an R-module, so that S ⊗R M is an S-module by extension of scalars. If M is a ﬂat R-module, show that S ⊗R M is a ﬂat S-module. 6. Let S be a multiplicative subset of the commutative ring R. Show that for any R-module M , S −1 R ⊗R M ∼ = S −1 M via α : (r/s) ⊗ x → rx/s with inverse β : x/s → (1/s) ⊗ x. 7. Continuing Problem 6, show that S −1 R is a ﬂat R-module. 10.9 Direct and Inverse Limits

page 23 of Chapter 10

If M is the direct sum of R-modules Mi , then R-homomorphisms fi : Mi → N can be lifted uniquely to an R-homomorphism f : M → N . The direct limit construction generalizes this idea. [In category theory, there is a further generalization called the colimit. The terminology is consistent because the direct sum is the coproduct in the category of modules.] 10.9.1 Direct Systems A directed set is a partially ordered set I such that given any i, j ∈ I, there exists k ∈ I such that i ≤ k and j ≤ k. A typical example is the collection of ﬁnite subsets of a set, ordered by inclusion. If A and B are arbitrary ﬁnite subsets, then both A and B are contained in the ﬁnite set A ∪ B. Now suppose I is a directed set and we have a collection of objects Ai , i ∈ I, in a category C. Assume that whenever i ≤ j, there is a morphism h(i, j) : Ai → Aj . Assume further that the h(i, j) are compatible in the sense that if i ≤ j ≤ k and we apply h(i, j) followed by h(j, k), we get h(i, k). We also require that for each i, h(i, i) is the identity on Ai . The collection of objects and morphisms is called a direct system. As an example, take the objects to be the ﬁnitely generated submodules of a module, and the morphisms to be the natural inclusion maps. In this case, the directed set coincides with the set of objects, and the partial ordering is inclusion. 10.9.2 Direct Limits Suppose that {Ai , h(i, j), i, j ∈ I} is a direct system. The direct limit of the system will consist of an object A and morphisms αi : Ai → A. Just as with coproducts, we want to lift morphisms fj : Aj → B to a unique f : A → B, that is, f αj = fj for all j ∈ I. But we require that the maps αj be compatible with the h(i, j), in other words, αj h(i, j) = αi whenever i ≤ j. A similar constraint is imposed on the fj , namely, fj h(i, j) = fi , i ≤ j. Thus the direct limit is an object A along with compatible morphisms αj : Aj → A such that given compatible morphisms fj : Aj → B, there is a unique morphism f : A → B such that f αj = fj for all j. Figure 10.9.1 summarizes the discussion.

A α

αj

i

fi Ai

B

fj

h(i,j)

Aj

Figure 10.9.1 As in Section 10.2, any two direct limits of a given direct system are isomorphic. If the ordering on I is the equality relation, then the only element j such that i ≤ j is i itself. Compatibility is automatic, and the direct limit reduces to a coproduct. A popular notation for the direct limit is A = lim Ai . →

The direct limit is sometimes called an inductive limit. 10.9.3 Examples 1. A coproduct is a direct limit, as discussed above. In particular, a direct sum of modules is a direct limit.

page 24 of Chapter 10

2. Any module is the direct limit of its ﬁnitely generated submodules. [Use the direct system indicated in (10.9.1).] 3. The algebraic closure of a ﬁeld F can be constructed (informally) by adjoining roots of all possible polynomials in F [X]; see (3.3.7). This suggests that the algebraic closure is the direct limit of the collection of all ﬁnite extensions of F . This can be proved with the aid of (3.3.9). In the category of modules, direct limits always exist. 10.9.4 Theorem If {Mi , h(i, j), i, j ∈ I} is a direct system of R-modules, then the direct limit of the system exists. Proof. Take M to be (⊕i Mi )/N , with N the submodule of the direct sum generated by all elements of the form βj h(i, j)xi − βi xi , i ≤ j, xi ∈ Mi (1) where βi is the inclusion map of Mi into the direct sum. Deﬁne αi : Mi → M by αi xi = βi xi + N. The αi are compatible, because αj h(i, j)xi = βj h(i, j)xi + N = βi xi + N = αi xi . Given compatible fi : Mi → B, we deﬁne f : M → B by f (βi xi + N ) = fi xi , the only possible choice. This forces f αi = fi , provided we show that f is well-deﬁned. But an element of N of the form (1) is mapped by our proposed f to fj h(i, j)xi − fi xi which is 0 by compatibility of the fi . Thus f maps everything in N to 0, and the result follows. ♣ 10.9.5 Inverse Systems and Inverse Limits Inverse limits are dual to direct limits. An inverse system is deﬁned as in (10.9.1), except that if i ≤ j, then h(i, j) maps “backwards” from Aj to Ai . If we apply h(j, k) followed by h(i, j), we get h(i, k); as before, h(i, i) is the identity on Ai . The inverse limit of the inverse system {Ai , h(i, j), i, j ∈ I} is an object A along with morphisms pi : A → Ai . As with products, we want to lift morphisms fi : B → Ai to a unique f : B → A. There is a compatibility requirement on the pi and fi : if i ≤ j, then h(i, j)pj = pi , and similarly h(i, j)fj = fi . Thus the inverse limit is an object A along with compatible morphisms pi : A → Ai such that given compatible morphisms fi : B → Ai , there is a unique morphism f : B → A such that pi f = fi for all i. See Figure 10.9.2.

A pj

pi fi Ai

B

fj

h(i,j) Figure 10.9.2

Aj

page 25 of Chapter 10

As in Section 10.2, the universal mapping property determines the inverse limit up to isomorphism. If the ordering on I is the equality relation, the the inverse limit reduces to a product. In category theory, the limit is a generalization of the inverse limit. A popular notation for the inverse limit is A = lim Ai . ←

The inverse limit is sometimes called a projective limit. We constructed the direct limit of a family of modules by forming a quotient of the direct sum. By duality, we expect that the inverse limit involves a submodule of the direct product. 10.9.6 Theorem If {Mi , h(i, j), i, j ∈ I} is an inverse system of R-modules, then the inverse limit of the system exists. Proof. We take M to be the set of all x = (xi , i ∈ I) in the direct product i Mi such that h(i, j)xj = xi whenever i ≤ j. Let pi be the restriction to M of the projection on the ith factor. Then h(i, j)pj x = h(i, j)xj = xi = pi x, so the pi are compatible. Given compatible fi : N → Mi , let f be the product of the fi , that is, f x = (fi x, i ∈ I). By compatibility, h(i, j)fj x = fi x for i ≤ j, so f maps i Mi into M . By deﬁnition of f we have pi f = fi , and the result follows. ♣ 10.9.7 Example Recall from Section 7.9 that a p-adic integer can be represented as a0 + a1 p + a2 p2 + · · ·, where the ai belong to {0, 1, . . . , p − 1}. If we discard all terms after ar−1 pr−1 , r = 1, 2, . . ., we get the ring Zpr . These rings form an inverse system; if x ∈ Zps and r ≤ s, we take h(r, s)x to be the residue of x mod pr . The inverse limit of this system is the ring of p-adic integers. Problems For Section 10.9 1. In Theorem (10.9.6), why can’t we say “obviously”, since direct limits exist in the category of modules, inverse limits must also exist by duality. 2. Show that in the category of modules over a commutative ring, the tensor product commutes with direct limits. In other words, lim (M ⊗ Ni ) = M ⊗ lim Ni →

→

assuming that the direct limit of the Ni exists. 3. For each n = 1, 2, . . ., let An be an R-module, with A1 ⊆ A2 ⊆ A3 ⊆ · · ·. Take h(i, j), i ≤ j, to be the inclusion map. What is the direct limit of the An ? (Be more explicit than in (10.9.4).) 4. Suppose that A, B and C are the direct limits of direct systems {Ai }, {Bi } and {Ci } of R-modules. Assume that for each i, the sequence

Ai

fi →

Bi

gi →

Ci

is exact. Give an intuitive argument to suggest that the sequence f A →

B

is exact. Thus direct limit is an exact functor.

g →

C

page 26 of Chapter 10

[A lot of formalism is being suppressed here. We must make the collection of direct systems into a category, and deﬁne a morphism in that category. This forces compatibility conditions on the fi and gi : fj hA (i, j) = hB (i, j)fi , gj hB (i, j) = hC (i, j)gi . The direct limit functor takes a direct system to its direct limit, but we must also specify what it does to morphisms.] A possible strategy is to claim that since an element of a direct sum has only ﬁnitely many nonzero components, exactness at B is equivalent to exactness at each Bi . This is unconvincing because the direct limit is not simply a direct sum, but a quotient of a direct sum. Suggestions are welcome! Problems 5 and 6 give some additional properties of direct products. 5. Show that

HomR (⊕Ai , B) ∼ =

HomR (Ai , B).

i

6. Show that HomR (A,

Bi ) ∼ =

i

HomR (A, Bi ).

i

7. If M is a nonzero R-module that is both projective and injective, where R is an integral domain that is not a ﬁeld, show that HomR (M, R) = 0. 8. Let R be an integral domain that is not a ﬁeld. If M is an R-module that is both projective and injective, show that M = 0. Appendix to Chapter 10 We have seen that an abelian group is injective if and only if it is divisible. In this appendix we give an explicit characterization of such groups. A10.1 Deﬁnitions and Comments Let G be an abelian group, and T the torsion subgroup of G (the elements of G of ﬁnite order). Then G/T is torsion-free, since n(x+T ) = 0 implies nx ∈ T , hence x ∈ T . If p is a ﬁxed prime, the primary component Gp associated with p consists of all elements whose order is a power of p. Note that Gp is a subgroup of G, for if pn a = pm b = 0, n ≥ m, then pn (a − b) = 0. (We use the fact that G is abelian; for example, 3(a − b) = a − b + a − b + a − b = a + a + a − b − b − b.) A10.2 Proposition The torsion subgroup T is the direct sum of the primary components Gp , p prime. k r Proof. Suppose x has order m = j=1 pj j . If mi = m/pri i , then the greatest common divisor of the miis 1, so there are integers a1 , . . . , ak such that a1 m1 + · · · + ak mk = 1. k Thus x = 1x = i=1 ai (mi x), and (by deﬁnition of mi ) mi x has order pri i and therefore belongs to the primary component Gpi . This proves that G is the sum of the Gp . To show that the sum is direct, assume 0 = x ∈ Gp ∩ q=p Gq . Then the order of x is a power of p and also a product of prime factors unequal to p, which is impossible. For example, if y has order 9 and z has order 125, then 9(125)(y + z) = 0, so the order of y + z is of the form 3r 5s . ♣ A10.3 Deﬁnitions and Comments A Pr¨ ufer group, also called a quasicyclic group and denoted by Z(p∞ ), is a p-primary component of Q/Z, the rationals mod 1. Since every element of Q/Z has ﬁnite order, it follows from (A10.2) that Z(p∞ ). Q/Z = p

Now an element of Q/Z whose order is a power of p must be of the form a/pr + Z for some integer a and nonnegative integer r. It follows that the elements ar = 1/pr + Z, r = 1, 2, . . ., generate Z(p∞ ). These elements satisfy the following relations: pa1 = 0, pa2 = a1 , . . . , par+1 = ar , . . .

page 27 of Chapter 10

A10.4 Proposition Let H be a group deﬁned by generators b1 , b2 , . . . and relations pb1 = 0, pb2 = b1 , . . . , pbr+1 = br , . . .. Then H is isomorphic to Z(p∞ ). Proof. First note that the relations imply that every element of H is an integer multiple of some bi . Here is a typical computation: 4b7 + 6b10 + 2b14 = 4(pb8 ) + 6(pb11 ) + 2b14 = · · · = 4(p7 b14 ) + 6(p4 b14 ) + 2b14 = (4p7 + 6p4 + 2)b14 . By (5.8.5), there is an epimorphism f : H → Z(p∞ ), and by the proof of (5.8.5), we can take f (bi ) = ai for all i. To show that f is injective, suppose f (cbi ) = 0 where c ∈ Z. Then cf (bi ) = cai = 0, so c/pi ∈ Z, in other words, pi divides c. ( We can reverse this argument to conclude that f (cbi ) = 0 iﬀ pi divides c.) But the relations imply that pi bi = 0, and since c is a multiple of pi , we have cbi = 0. ♣ A10.5 Proposition Let G be a divisible abelian group. Then its torsion subgroup T is also divisible. Moreover, G can be written as T ⊕ D, where D is torsion-free and divisible. Proof. If x ∈ T and 0 = n ∈ Z, then for some y ∈ G we have ny = x. Thus in the torsion-free group G/T we have n(y + T ) = x + T = 0. But then ny ∈ T , so (as in (A10.1)) y ∈ T and T is divisible, hence injective by (10.6.6). By (10.6.2), the exact sequence 0 → T → G → G/T → 0 splits, so G ∼ = T ⊕ G/T . Since G/T is torsion-free and divisible (see (10.6.5)), the result follows. ♣ We are going to show that an abelian group is divisible iﬀ it is a direct sum of copies of Q (the additive group of rationals) and quasicyclic groups. To show that every divisible abelian group has this form, it suﬃces, by (A10.2), (A10.5) and the fact that a direct sum of divisible abelian groups is divisible, to consider only two cases, G torsion-free and G a p-group. A10.6 Proposition If G is a divisible, torsion-free abelian group, then G is isomorphic to a direct sum of copies of Q. Proof. The result follows from the observation that G can be regarded as a Q-module, that is, a vector space over Q; see Section 10.7, Problem 7. ♣ For any abelian group G, let G[n] = {x ∈ G : nx = 0}. A10.7 Proposition Let G and H be divisible abelian p-groups. Then any isomorphism ϕ of G[p] and H[p] can be extended to an isomorphism ψ of G and H. Proof. Our candidate ψ arises from the injectivity of H, as the diagram below indicates. H ↑ 0

→

G[p]

ψ →

G

The map from G[p] to G is inclusion, and the map from G[p] to H is the composition of ϕ and the inclusion from H[p] to H. Suppose that x ∈ G and the order of x is |x| = pn . We will prove by induction that ψ(x) = 0 implies x = 0. If n = 1, then x ∈ G[p], so ψ(x) = ϕ(x), and the result follows because ϕ is injective. For the inductive step, suppose |x| = pn+1 and ψ(x) = 0. Then |px| = pn and ψ(px) = pψ(x) = 0. By induction hypothesis, px = 0, which contradicts the assumption that x has order pn+1 . Now we prove by induction that ψ is surjective. Explicitly, if y ∈ H and |y| = pn , then y belongs to the image of ψ. If n = 1, then y ∈ H[p] and the result follows because ϕ is

page 28 of Chapter 10

surjective. If |y| = pn+1 , then pn y ∈ H[p], so for some x ∈ G[p] we have ϕ(x) = pn y. Since G is divisible, there exists g ∈ G such that pn g = x. Then pn (y − ψ(g)) = pn y − ψ(pn g) = pn y − ψ(x) = pn y − ϕ(x) = 0. By induction hypothesis, there is an element z ∈ G such that ψ(z) = y − ψ(g). Thus ψ(g + z) = y. ♣ A10.8 Theorem An abelian group G is divisible if and only if G is a direct sum of copies of Q and quasicyclic groups. Proof. Suppose that G is such a direct sum. Since Q and Z(p∞ ) are divisible [Z(p∞ ) is a direct summand of the divisible group Q/Z], and a direct sum of divisible abelian groups is divisible, G must be divisible. Conversely, assume G divisible. In view of (A10.6) and the discussion preceding it, we may assume that G is a p-group. But then G[p] is a vector space over the ﬁeld Fp = Z/pZ; the scalar multiplication is given by (n + pZ)g = ng. Since pg = 0, scalar multiplication is well-deﬁned. If the dimension of G[p] over Fp is d, let H be the direct sum of d copies of Z(p∞ ). An element of order p in a component of the direct sum is an integer multiple of 1/p + Z, and consequently H[p] is also a d-dimensional vector space over Fp . Thus G[p] is isomorphic to H[p], and it follows from (A10.7) that G is isomorphic to H. ♣

page 1 of Supplement

SUPPLEMENT: THE LONG EXACT HOMOLOGY SEQUENCE AND APPLICATIONS S1. Chain Complexes In the supplement, we will develop some of the building blocks for algebraic topology. As we go along, we will make brief comments [in brackets] indicating the connection between the algebraic machinery and the topological setting, but for best results here, please consult a text or attend lectures on algebraic topology. S1.1 Deﬁnitions and Comments A chain complex (or simply a complex) C∗ is a family of R-modules Cn , n ∈ Z, along with R-homomorphisms dn : Cn → Cn−1 called diﬀerentials, satisfying dn dn+1 = 0 for all n. A chain complex with only ﬁnitely many Cn ’s is allowed; it can always be extended with the aid of zero modules and zero maps. [In topology, Cn is the abelian group of n-chains, that is, all formal linear combinations with integer coeﬃcients of n-simplices in a topological space X. The map dn is the boundary operator, which assigns to an n-simplex an n − 1-chain that represents the oriented boundary of the simplex.] The kernel of dn is written Zn (C∗ ) or just Zn ; elements of Zn are called cycles in dimension n. The image of dn+1 is written Bn (C∗ ) or just Bn ; elements of Bn are called boundaries in dimension n. Since the composition of two successive diﬀerentials is 0, it follows that Bn ⊆ Zn . The quotient Zn /Bn is written Hn (C∗ ) or just Hn ; it is called the nth homology module (or homology group if the underlying ring R is Z). [The key idea of algebraic topology is the association of an algebraic object, the collection of homology groups Hn (X), to a topological space X. If two spaces X and Y are homeomorphic, in fact if they merely have the same homotopy type, then Hn (X) and Hn (Y ) are isomorphic for all n. Thus the homology groups can be used to distinguish between topological spaces; if the homology groups diﬀer, the spaces cannot be homeomorphic.] Note that any exact sequence is a complex, since the composition of successive maps is 0. S1.2 Deﬁnition A chain map f : C∗ → D∗ from a chain complex C∗ to a chain complex D∗ is a collection of module homomorphisms fn : Cn → Dn , such that for all n, the following diagram is commutative. fn Cn → Dn dn ↓ Cn−1

↓ dn → fn−1

Dn−1

We use the same symbol dn to refer to the diﬀerentials in C∗ and D∗ . [If f : X → Y is a continuous map of topological spaces and σ is a singular n-simplex in X, then f# (σ) = f ◦ σ is a singular n-simplex in Y , and f# extends to a homomorphism of n-chains. If we assemble the f# ’s for n = 0, 1, . . ., the result is a chain map.] S1.3 Proposition A chain map f takes cycles to cycles and boundaries to boundaries. Consequently, the map zn + Bn (C∗ ) → fn (zn ) + Bn (D∗ ) is a well-deﬁned homomorphism from Hn (C∗ ) to Hn (D∗ ). It is denoted by Hn (f ). Proof. If z ∈ Zn (C∗ ), then since f is a chain map, dn fn (z) = fn−1 dn (z) = fn−1 (0) = 0. Therefore fn (z) ∈ Zn (D∗ ). If b ∈ Bn (C∗ ), then dn+1 c = b for some c ∈ Cn+1 . Then fn (b) = fn (dn+1 c) = dn+1 fn+1 c, so fn (b) ∈ Bn (D∗ ). ♣ S1.4 The Homology Functors We can create a category whose objects are chain complexes and whose morphisms are chain maps. The composition gf of two chain maps

page 2 of Supplement

f : C∗ → D∗ and g : D∗ → E∗ is the collection of homomorphisms gn fn , n ∈ Z. For any n, we associate with the chain complex C∗ its nth homology module Hn (C∗ ), and we associate with the chain map f : C∗ → D∗ the map Hn (f ) : Hn (C∗ ) → Hn (D∗ ) deﬁned in (S1.3). Since Hn (gf ) = Hn (g)Hn (f ) and Hn (1C∗ ) is the identity on Hn (C∗ ), Hn is a functor, called the nth homology functor. S1.5 Chain Homotopy Let f and g be chain maps from C∗ to D∗ . We say that f and g are chain homotopic and write f g if there exist homomorphisms hn : Cn → Dn+1 such that fn − gn = dn+1 hn + hn−1 dn ; see the diagram below.

dn

C

C n-1

n

hn f -g n n

D n+1

h n-1

Dn

d n+1

[If f and g are homotopic maps from a topological space X to a topological space Y , then the maps f# and g# (see the discussion in (S1.2)) are chain homotopic,] S1.6 Proposition If f and g are chain homotopic, then Hn (f ) = Hn (g). Proof. Let z ∈ Zn (C∗ ). Then fn (z) − gn (z) = (dn+1 hn + hn−1 dn )z ∈ Bn (D∗ ) since dn z = 0. Thus fn (z) + Bn (D∗ ) = gn (z) + Bn (D∗ ), in other words, Hn (f ) = Hn (g). ♣ S2. The Snake Lemma We isolate the main ingredient of the long exact homology sequence. After an elaborate diagram chase, a homomorphism between two modules is constructed. The domain and codomain of the homomorphism are far apart in the diagram, and the arrow joining them tends to wiggle like a serpent. First, a result about kernels and cokernels of module homomorphisms. S2.1 Lemma Assume that the diagram below is commutative. A

f →

d↓ C

B ↓e

→ g

D

(i) f induces a homomorphism on kernels, that is, f (ker d) ⊆ ker e. (ii) g induces a homomorphism on cokernels, that is, the map y +im d → g(y)+im e, y ∈ C, is a well-deﬁned homomorphism from coker d to coker e. (iii) If f is injective, so is the map induced by f , and if g is surjective, so is the map induced by g. Proof. (i) If x ∈ A and d(x) = 0, then ef (x) = gd(x) = g0 = 0.

page 3 of Supplement

(ii) If y ∈ im d, then y = dx for some x ∈ A. Thus gy = gdx = ef x ∈ im e. Since g is a homomorphism, the induced map is also. (iii) The ﬁrst statement holds because the map induced by f is simply a restriction. The second statement follows from the form of the map induced by g. ♣ Now refer to our snake diagram, Figure S2.1. Initially, we are given only the second and third rows (ABE0 and 0CDF), along with the maps d, e and h. Commutativity of the squares ABDC and BEFD is assumed, along with exactness of the rows. The diagram is now enlarged as follows. Take A =ker d, and let the map from A to A be inclusion. Take C = coker d, and let the map from C to C be canonical. Augment columns 2 and 3 in a similar fashion. Let A → B be the map induced by f on kernels, and let C → D be the map induced by g on cokernels. Similarly, add B → E and D → F . The enlarged diagram is commutative by (S2.1), and it has exact columns by construction.

A’

B’

f

A

s B

0

E

e

d g 0

E’

h t

C

D

C’

D’

F

F’

Figure S2.1 S2.2 Lemma The ﬁrst and fourth rows of the enlarged snake diagram are exact. Proof. This is an instructive diagram chase, showing many standard patterns. Induced maps will be denoted by an overbar, and we ﬁrst prove exactness at B . If x ∈ A = ker d and y = f x = f x, then sy = sf x = 0, so y ∈ker s. On the other hand, if y ∈ B ⊆ B and sy = sy = 0, then y = f x for some x ∈ A. Thus 0 = ey = ef x = gdx, and since g is injective, dx = 0. Therefore y = f x with x ∈ A , and y ∈im f . Now to prove exactness at D , let x ∈ C. Then t(gx+im e)= tgx+im h = 0 by exactness of the third row, so im g ⊆ker t. Conversely, if y ∈ D and t(y + im e) = ty + im h = 0, then ty = hz for some z ∈ E. Since s is surjective, z = sx for some x ∈ B. Now ty = hz = hsx = tex so y − ex ∈ ker t = im g, say y − ex = gw, w ∈ C. Therefore y + im e = g(w + im d) and y + im e ∈ im g. ♣ S2.3 Remark Sometimes an even bigger snake diagram is given, with column 1 assumed to be an exact sequence 0

→ A

→

A

d → C

→

C

→

0

page 4 of Supplement

and similarly for columns 2 and 3. This is nothing new, because by replacing modules by isomorphic copies we can assume that A is the kernel of d, C is the cokernel of d, A → A is inclusion, and C → C is canonical. S2.4 The Connecting Homomorphism We will now connect E to C in the snake diagram while preserving exactness. The idea is to zig-zag through the diagram along the path E EBDCC . Let z ∈ E ⊆ E; Since s is surjective, there exists y ∈ B such that z = sy. Then tey = hsy = hz = 0 since E = ker h. Thus ey ∈ ker t = im g, so ey = gx for some x ∈ C. We deﬁne the connecting homomorphism ∂ : E → C by ∂z = x + im d. Symbolically, ∂ = [g −1 ◦ e ◦ s−1 ] where the brackets indicate that ∂z is the coset of x in C = C/im d. We must show that ∂ is well-deﬁned. Suppose that y is another element of B with sy = z. Then y − y ∈ ker s = im f , so y − y = f u for some u ∈ A. Thus e(y − y ) = ef u = gdu. Now we know from the above computation that ey = gx for some x ∈ C, and similarly ey = gx for some x ∈ C. Therefore g(x − x ) = gdu, so x − x − du ∈ ker g. Since g is injective, x − x = du, so x + im d = x + im d. Thus ∂z is independent of the choice of the representatives y and x. Since every map in the diagram is a homomorphism, so is ∂.

S2.5 Snake Lemma The sequence A

f →

B

s →

E

∂ →

C

g →

D

t →

F

is exact. Proof. In view of (S2.2), we need only show exactness at E and C . If z = sy, y ∈ B = ker e, then ey = 0, so ∂z = 0 by deﬁnition of ∂. Thus im s ⊆ ker ∂. Conversely, assume ∂z = 0, and let x and y be as in the deﬁnition of ∂. Then x = du for some u ∈ A, hence gx = gdu = ef u. But gx = ey by deﬁnition of ∂, so y − f u ∈ ker e = B . Since z = sy by deﬁnition of ∂, we have z = s(y − f u + f u) = s(y − f u) ∈ im s. To show exactness at C , consider an element ∂z in the image of ∂. Then ∂z = x + im d , so g∂z = gx + im e. But gx = ey by deﬁnition of ∂, so g∂z = 0 and ∂z ∈ ker g. Conversely, suppose x ∈ C and g(x + im d) = gx + im e = 0. Then gx = ey for some y ∈ B. If z = sy, then hsy = tey = tgx = 0 by exactness of the third row. Thus z ∈ E and (by deﬁnition of ∂) we have ∂z = x + im d. Consequently, x + im d ∈ im ∂. ♣ S3. The Long Exact Homology Sequence S3.1 Deﬁnition We say that

0

→

C∗

f →

D∗

g →

E∗

→

0

where f and g are chain maps, is a short exact sequence of chain complexes if for each n, the corresponding sequence formed by the component maps fn : Cn → Dn and gn : Dn → En , is short exact. We will construct connecting homomorphisms ∂n : Hn (E∗ ) → Hn−1 (C∗ ) such that the sequence ···

g ∂ → Hn+1 (E∗ ) →

f Hn (C∗ ) →

g Hn (D∗ ) →

∂ f Hn (E∗ ) → Hn−1 (C∗ ) →

···

page 5 of Supplement

is exact. [We have taken some liberties with the notation. In the second diagram, f stands for the map induced by fn on homology, namely, Hn (f ); similarly for g.] The second diagram is the long exact homology sequence, and the result may be summarized as follows. S3.2 Theorem A short exact sequence of chain complexes induces a long exact sequence of homology modules. Proof. This is a double application of the snake lemma. The main ingredient is the following snake diagram. Cn /Bn (C∗ ) → Dn /Bn (D∗ ) → d↓ 0 →

En /Bn (E∗ ) →

d↓ →

Zn−1 (C∗ )

0

d↓ →

Zn−1 (D∗ )

Zn−1 (E∗ )

The horizontal maps are derived from the chain maps f and g, and the vertical maps are given by d(xn + Bn ) = dxn . The kernel of a vertical map is {xn + Bn : xn ∈ Zn } = Hn , and the cokernel is Zn−1 /Bn−1 = Hn−1 . The diagram is commutative by the deﬁnition of a chain map. But in order to apply the snake lemma, we must verify that the rows are exact, and this involves another application of the snake lemma. The appropriate diagram is 0

→

→

Cn d↓

0 →

Dn

→

En

d↓

Cn−1

→

Dn−1

→

0

→

0

d↓ →

En−1

where the horizontal maps are again derived from f and g. The exactness of the rows of the ﬁrst diagram follows from (S2.2) and part (iii) of (S2.1), shifting indices from n to n ± 1 as needed. ♣ S3.3 The connecting homomorphism explicitly If z ∈ Hn (E∗ ), then z = zn + Bn (E∗ ) for some zn ∈ Zn (E∗ ). We apply (S2.4) to compute ∂z. We have zn + Bn (E∗ ) = gn (yn + Bn (D∗ )) for some yn ∈ Dn . Then dyn ∈ Zn−1 (D∗ ) and dyn = fn−1 (xn−1 ) for some xn−1 ∈ Zn−1 (C∗ ). Finally, ∂z = xn−1 + Bn−1 (C∗ ). S3.4 Naturality Suppose that we have a commutative diagram of short exact sequences of chain complexes, as shown below. 0 →

→

C ↓

0 →

D

→

E

↓

C

→

D

→

0

→

0

↓ E

→

Then there is a corresponding commutative diagram of long exact sequences: ∂ · · · → Hn (C∗ ) → ↓ · · · → Hn (C∗ ) → ∂

Hn (D∗ ) → ↓ Hn (D∗ ) →

∂ Hn (E∗ ) → Hn−1 (C∗ ) → ↓

···

↓

Hn (E∗ ) → Hn−1 (C∗ ) → ∂

···

Proof. The homology functor, indeed any functor, preserves commutative diagrams, so the two squares on the left commute. For the third square, an informal argument may help to

page 6 of Supplement

illuminate the idea. Trace through the explicit construction of ∂ in (S3.3), and let f be the vertical chain map in the commutative diagram of short exact sequences. The ﬁrst step in the process is zn + Bn (E∗ ) → yn + Bn (D∗ ). By commutativity,

f zn + Bn (E∗ ) → f yn + Bn (D∗ ).

Continuing in this fashion, we ﬁnd that if ∂z = xn−1 + Bn−1 (C∗ ), then ∂(f z) = f xn−1 + Bn−1 (C∗ ) = f (xn−1 + Bn−1 (C∗ )) = f (∂z). ♣ A formal proof can be found in ”An Introduction to Algebraic Topology” by J. Rotman, page 95. S4. Projective and Injective Resolutions The functors Tor and Ext are developed with the aid of projective and injective resolutions of a module, and we will now examine these constructions. S4.1 Deﬁnitions and Comments A left resolution of a module M is an exact sequence · · · P2 → P1 → P0 → M → 0. A left resolution is a projective resolution if every Pi is projective, a free resolution if every Pi is free. By the ﬁrst isomorphism theorem, M is isomorphic to the cokernel of the map P1 → P0 , so in a sense no information is lost if M is removed. A deleted projective resolution is of the form · · · P 2 → P1 → P0 → 0 ↓ M and the deleted version turns out to be more convenient in computations. Notice that in a deleted projective resolution, exactness at P0 no longer holds because the map P1 → P0 need not be surjective. Resolutions with only ﬁnitely many Pn ’s are allowed, provided that the module on the extreme left is 0. The sequence can then be extended via zero modules and zero maps. Dually, a right resolution of M is an exact sequence 0 → M → E0 → E1 → E2 · · · ; we have an injective resolution if every Ei is injective, . A deleted injective resolution has the form 0 → E0 → E1 → E2 · · · ↑ M Exactness at E0 no longer holds because the map E0 → E1 need not be injective. We will use the notation P∗ → M for a projective resolution, and M → E∗ for an injective resolution. S4.2 Proposition Every module M has a free (hence projective) resolution. Proof. By (4.3.6), M is a homomorphic image of a free module F0 . Let K0 be the kernel of the map from F0 onto M . In turn, there is a homomorphism with kernel K1 from a free module F1 onto K0 , and we have the following diagram: 0 → K 1 → F1 → K 0 → F 0 → M → 0

page 7 of Supplement

Composing the maps F1 → K0 and K0 → F0 , we get 0 → K 1 → F1 → F 0 → M → 0 which is exact. But now we can ﬁnd a free module F2 and a homomorphism with kernel K2 mapping F2 onto K1 . The above process can be iterated to produce the desired free resolution. ♣ Specifying a module by generators and relations (see (4.6.6) for abelian groups) involves ﬁnding an appropriate F0 and K0 , as in the ﬁrst step of the above iterative process. Thus a projective resolution may be regarded as a generalization of a speciﬁcation by generators and relations. Injective resolutions can be handled by dualizing the proof of (S4.2). S4.3 Proposition Every module M has an injective resolution. Proof. By (10.7.4), M can be embedded in an injective module E0 . Let C0 be the cokernel of M → E0 , and map E0 canonically onto C0 . Embed C0 in an injective module E1 , and let C1 be the cokernel of the embedding map. We have the following diagram: 0 → M → E0 → C0 → E1 → C1 → 0 Composing E0 → C0 and C0 → E1 , we have 0 → M → E0 → E1 → C1 → 0 which is exact. Iterate to produce the desired injective resolution. ♣ S5. Derived Functors S5.1 Left Derived Functors Suppose that F is a right exact functor from modules to modules. (In general, the domain and codomain of F can be abelian categories, but the example we have in mind is M ⊗R .) Given a short exact sequence 0 → A → B → C → 0, we form deleted projective resolutions PA∗ → A, PB∗ → B, PC∗ → C. It is shown in texts on homological algebra that it is possible to deﬁne chain maps to produce a short exact sequence of complexes as shown below. 0 →

A

→

↑ 0 →

PA∗

B

→

↑ →

PB∗

C

→

0

→

0

↑ →

PC∗

The functor F will preserve exactness in the diagram, except at the top row, where we only have F A → F B → F C → 0 exact. But remember that we are using deleted resolutions, so that the ﬁrst row is suppressed. The left derived functors of F are deﬁned by taking the homology of the complex F (P ), that is, (Ln F )(A) = Hn [F (PA∗ )]. The word “left” is used because the Ln F are computed using left resolutions. It can be shown that up to natural equivalence, the derived functors are independent of the particular projective resolutions chosen. By (S3.2), we have the following long exact sequence: ∂ · · · → (Ln F )(A) → (Ln F )(B) →

∂ (Ln F )(C) → (Ln−1 F )(A) →

···

page 8 of Supplement

S5.2 Right Derived Functors Suppose now that F is a left exact functor from modules to modules, e.g., HomR (M, ). We can dualize the discussion in (S5.1) by reversing the vertical arrows in the commutative diagram of complexes, and replacing projective resolutions such as PA∗ by injective resolutions EA∗ . The right derived functors of F are deﬁned by taking the homology of F (E). In other words, (Rn F )(A) = H n [F (EA∗ )] where the superscript n indicates that we are using right resolutions and the indices are increasing as we move away from the starting point. By (S3.2), we have the following long exact sequence: ∂ · · · → (Rn F )(A) →

(Rn F )(B) →

∂ (Rn F )(C) → Rn+1 F (A) →

···

S5.3 Lemma (L0 F )(A) ∼ = F (A) ∼ = (R0 F )(A) Proof. This is a good illustration of the advantage of deleted resolutions. If P∗ → A, we have the following diagram: F (P1 ) →

F (P0 ) → 0 ↓ F (A) ↓ 0

The kernel of F (P0 ) → 0 is F (P0 ), so the 0th homology module (L0 F )(A) is F (P0 ) mod the image of F (P1 ) → F (P0 ) [=the kernel of F (P0 ) → F (A).] By the ﬁrst isomorphism theorem and the right exactness of F , (L0 F )(A) ∼ = F (A). To establish the other isomorphism, we switch to injective resolutions and reverse arrows: 0

→ F (E0 ) → ↑ F (A) ↑ 0

F (E1 )

The kernel of F (E0 ) → F (E1 ) is isomorphic to F (A) by left exactness of F , and the image of 0 → F (E0 ) is 0. Thus (R0 F )(A) ∼ = F (A). ♣ S5.4 Lemma If A is projective, then (Ln F )(A) = 0 for every n > 0; if A is injective, then (Rn F )(A) = 0 for every n > 0. Proof. If A is projective [resp. injective], then 0 → A → A → 0 is a projective [resp. injective] resolution of A. Switching to a deleted resolution, we have 0 → A → 0 in each case, and the result follows. ♣ S5.5 Deﬁnitions and Comments If F is the right exact functor M ⊗R , the left derived functor Ln F is called TorR ). If F is the left exact functor HomR (M, ), the right n (M, derived functor Rn F is called ExtnR (M, ). It can be shown that the Ext functors can also be computed using projective resolutions and the contravariant hom functor. Speciﬁcally, ExtnR (M, N ) = [Rn HomR ( , N )](M ). A switch from injective to projective resolutions is a simpliﬁcation, because projective resolutions are easier to ﬁnd in practice.

page 9 of Supplement

The next three results sharpen Lemma S5.4. [The ring R is assumed ﬁxed, and we write ⊗R simply as ⊗. Similarly, we drop the R in TorR and ExtR . When discussing Tor, we assume R commutative.] S5.6 Proposition If M is an R-module, the following conditions are equivalent. (i) M is ﬂat; (ii) Torn (M, N ) = 0 for all n ≥ 1 and all modules N ; (iii) Tor1 (M, N ) = 0 for all modules N . Proof. (i) implies (ii): Let P∗ → N be a projective resolution of N . Since M ⊗ functor (see (10.8.1)), the sequence

is an exact

· · · → M ⊗ P1 → M ⊗ P0 → M ⊗ N → 0 is exact. Switching to a deleted resolution, we have exactness up to M ⊗ P1 but not at M ⊗ P0 . Since the homology modules derived from an exact sequence are 0, the result follows. (ii) implies (iii): Take n = 1. (iii) implies (i): If 0 → A → B → C → 0 is a short exact sequence, then by (S5.1), we have the following long exact sequence: · · · Tor1 (M, C) → Tor0 (M, A) → Tor0 (M, B) → Tor0 (M, C) → 0. By hypothesis, Tor1 (M, C) = 0, so by (S5.3), 0→M ⊗A→M ⊗B →M ⊗C →0 is exact, and therefore M is ﬂat. ♣ S5.7 Proposition If M is an R-module, the following conditions are equivalent. (i) M is projective; (ii) Extn (M, N ) = 0 for all n ≥ 1 and all modules N ; (iii) Ext1 (M, N ) = 0 for all modules N . Proof. (i) implies (ii): By (S5.4) and (S5.5), Extn (M, N ) = [Extn ( , N )](M ) = 0 for n ≥ 1. (ii) implies (iii): Take n = 1. (iii) implies (i): Let 0 → A → B → M → 0 be a short exact sequence. If N is any module, then using projective resolutions and the contravariant hom functor to construct Ext, as in (S5.5), we get the following long exact sequence: 0 → Ext0 (M, N ) → Ext0 (B, N ) → Ext0 (A, N ) → Ext1 (M, N ) → · · · By (iii) and (S5.3), 0 → Hom(M, N ) → Hom(B, N ) → Hom(A, N ) → 0 is exact. Take N = A and let g be the map from A to B. Then the map g ∗ from Hom(B, A) to Hom(A, A) is surjective. But 1A ∈ Hom(A, A), so there is a homomorphism f : B → A such that g ∗ (f ) = f g = 1A . Therefore the sequence 0 → A → B → M → 0 splits, so by (10.5.3), M is projective. ♣

page 10 of Supplement

S5.8 Corollary If N is an R-module, the following conditions are equivalent. (a) N is injective; (b) Extn (M, N ) = 0 for all n ≥ 1 and all modules M ; (c) Ext1 (M, N ) = 0 for all modules M . Proof. Simply saying “duality” may be unconvincing, so let’s give some details. For (a) implies (b), we have [Extn (M, )](N ) = 0. For (c) implies (a), note that the exact sequence 0 → N → A → B → 0 induces the exact sequence 0 → Ext0 (M, N ) → Ext0 (M, A) → Ext0 (M, B) → 0. Replace Ext0 (M, N ) by Hom(M, N ), and similarly for the other terms. Then take M = B and proceed exactly as in (S5.7). ♣ S6. Some Properties of Ext and Tor We will compute ExtnR (A, B) and TorR n (A, B) in several interesting cases. S6.1 Example We will calculate ExtnZ (Zm , B) for an arbitrary abelian group B. To ease the notational burden slightly, we will omit the subscript Z in Ext and Hom, and use = (most of the time) when we really mean ∼ =. We have the following projective resolution of Zm : m 0 → Z → Z → Zm → 0 where the m over the arrow indicates multiplication by m. Switching to a deleted resolution and applying the contravariant hom functor, we get 0 →

Hom(Z, B)

m →

Hom(Z, B) → 0

↑ Hom(Zm , B) But by (9.4.1), we have

HomR (R, B) ∼ =B

(1)

and the above diagram becomes 0

→

B

m →

B

→

0

(2)

By (S5.3), Ext0 (Zm , B) = Hom(Zm , B). Now a homomorphism f from Zm to B is determined by f (1), and f (m) = mf (1) = 0. If B(m) is the set of all elements of B that are annihilated by m, that is, B(m) = {x ∈ B : mx = 0}, then the map of B(m) to Hom(Zm , B) given by x → f where f (1) = x, is an isomorphism. Thus Ext0 (Zm , B) = B(m). It follows from (2) that Extn (Zm , B) = 0, n ≥ 2 and Ext1 (Zm , B) = ker(B → 0)/im(B → B) = B/mB. The computation for n ≥ 2 is a special case of a more general result.

page 11 of Supplement

S6.2 Proposition ExtnZ (A, B) = 0 for all n ≥ 2 and all abelian groups A and B. Proof. If B is embedded in an injective module E, we have the exact sequence 0 → B → E → E/B → 0. This is an injective resolution of B since E/B is divisible, hence injective; see (10.6.5) and (10.6.6). Applying the functor Hom(A, ) = HomZ (A, ) and switching to a deleted resolution, we get the sequence 0 →

Hom(A, E)

→

Hom(A, E/B) →

0

↑ Hom(A, B) whose homology is 0 for all n ≥ 2. ♣ S6.3 Lemma Ext0Z (Z, B) = HomZ (Z, B) = B and Ext1Z (Z, B) = 0. Proof. The ﬁrst equality follows from (S5.3) and the second from (1) of (S6.1). Since Z is projective, the last statement follows from (S5.7). ♣ S6.4 Example We will compute TorZn (Zm , B) for an arbitrary abelian group B. As before, we drop the superscript Z and write = for ∼ =. We use the same projective resolution of Zm as in (S6.1), and apply the functor ⊗ B. Since R ⊗R B ∼ = B by (8.7.6), we reach diagram (2) as before. Thus Torn (Zm , B) = 0, n ≥ 2; Tor1 (Zm , B) = ker(B → B) = {x ∈ B : mx = 0} = B(m); Tor0 (Zm , B) = Zm ⊗ B = B/mB. [To verify the last equality, use the universal mapping property of the tensor product to produce a map of Zm ⊗ B to B/mB such that n ⊗ x → n(x + mB). The inverse of this map is x + mB → 1 ⊗ x.] The result for n ≥ 2 generalizes as in (S6.2): S6.5 Proposition TorZn (A, B) = 0 for all n ≥ 2 and all abelian groups A and B. Proof. B is the homomorphic image of a free module F . If K is the kernel of the homomorphism, then the exact sequence 0 → K → F → B → 0 is a free resolution of B. [K is a submodule of a free module over a PID, hence is free.] Switching to a deleted resolution and applying the tensor functor, we get a four term sequence as in (S6.2), and the homology must be 0 for n ≥ 2. ♣ S6.6 Lemma Tor1 (Z, B) = Tor1 (A, Z) = 0; Tor0 (Z, B) = Z ⊗ B = B. Proof. The ﬁrst two equalities follow from (S5.6) since Z is ﬂat. The other two equalities follow from (S5.3) and (8.7.6). ♣ S6.7 Finitely generated abelian groups We will show how to compute Extn (A, B) and Torn (A, B) for arbitrary ﬁnitely generated abelian groups A and B. By (4.6.3), A and B can be expressed as a ﬁnite direct sum of cyclic groups. Now Tor commutes with direct sums: Torn (A, ⊕rj=1 Bj ) = ⊕rj=1 Torn (A, Bj ). [The point is that if Pj∗ is a projective resolution of Bj , then the direct sum of the Pj∗ is a projective resolution of ⊕j Bj , by (10.5.4). Since the tensor functor is additive on direct sums, by (8.8.6(b)), the Tor functor will be additive as well. Similar results hold when the

page 12 of Supplement

direct sum is in the ﬁrst coordinate, and when Tor is replaced by Ext. (We use (10.6.3) and Problems 5 and 6 of Section 10.9, and note that the direct product and the direct sum are isomorphic when there are only ﬁnitely many factors.)] Thus to complete the computation, we need to know Ext(A, B) and Tor(A, B) when A = Z or Zm and B = Z or Zn . We have already done most of the work. By (S6.2) and (S6.5), Extn and Torn are identically 0 for n ≥ 2. By (S6.1), Ext0 (Zm , Z) = Z(m) = {x ∈ Z : mx = 0} = 0; Ext0 (Zm , Zn ) = Zn (m) = {x ∈ Zn : mx = 0} = Zd where d is the greatest common divisor of m and n. [For example, Z12 (8) = {0, 3, 6, 9} ∼ = Z4 . The point is that the product of two integers is their greatest common divisor times their least common multiple.] By (S6.3), Ext0 (Z, Z) = Hom(Z, Z) = Z; Ext0 (Z, Zn ) = Zn . By (S6.1), Ext1 (Zm , Z) = Z/mZ = Zm ; Ext1 (Zm , Zn ) = Zn /mZn = Zd as above. By (S5.7), Ext1 (Z, Z) = Ext1 (Z, Zn ) = 0 By (S6.4), Tor1 (Zm , Z) = Tor1 (Z, Zm ) = Z(m) = 0; Tor1 (Zm , Zn ) = Zn (m) = Zd . By (8.7.6) and (S6.4), Tor0 (Z, Z) = Z; Tor0 (Zm , Z) = Z/mZ = Zm ; Tor0 (Zm , Zn ) = Zn /mZn = Zd . Notice that Tor1 (A, B) is a torsion group for all ﬁnitely generated abelian groups A and B. This is a partial explanation of the term “Tor”. The Ext functor arises in the study of group extensions. S7 Base Change in the Tensor Product Let M be an A-module, and suppose that we have a ring homomorphism from A to B (all rings are assumed commutative). Then B ⊗A M becomes a B module (hence an A-module) via b(b ⊗ m) = bb ⊗ m. This is an example of base change, as discussed in (10.8.8). We examine some frequently occurring cases. First, consider B = A/I, where I is an ideal of A. S7.1 Proposition (A/I) ⊗A M ∼ = M/IM . Proof. Apply the (right exact) tensor functor to the exact sequence of A-modules 0 → I → A → A/I → 0 to get the exact sequence I ⊗A M → A ⊗A M → (A/I) ⊗A M → 0.

page 13 of Supplement

Recall from (8.7.6) that A ⊗A M is isomorphic to M via a ⊗ m → am. By the ﬁrst isomorphism theorem, (A/I) ⊗A M is isomorphic to M mod theimage of the map from I ⊗A M to M . This image is the collection of all ﬁnite sums ai mi with ai ∈ I and mi ∈ M , which is IM . ♣ Now consider B = S −1 A, where S is a multiplicative subset of A. S7.2 Proposition (S −1 A) ⊗A M ∼ = S −1 M . Proof. The map from S −1 A×M to S −1 M given by (a/s, m) → am/s is A-bilinear, so by the universal mapping property of the tensor product, there is a linear map α : S −1 A ⊗A M → S −1 M such that α((a/s) ⊗ m) = am/s. The inverse map β is given by β(m/s) = (1/s) ⊗ m. To show that β is well-deﬁned, suppose that m/s = m /s . Then for some t ∈ S we have ts m = tsm . Thus 1/s ⊗ m = ts /tss ⊗ m = 1/tss ⊗ ts m = 1/tss ⊗ tsm = 1/s ⊗ m . Now α followed by β takes a/s ⊗ m to am/s and then to 1/s ⊗ am = a/s ⊗ m. On the other hand, β followed by α takes m/s to 1/s ⊗ m and then to m/s. Consequently, α and β are inverses of each other and yield the desired isomorphism of S −1 A ⊗A M and S −1 M . ♣ Finally, we look at B = A[X]. S7.3 Proposition A[X] ⊗A M ∼ = M [X] where the elements of M [X] are of the form a0 m0 +a1 Xm1 +a2 X 2 m2 +· · ·+an X n mn , ai ∈ A, mi ∈ M, n = 0, 1, . . .. Proof. This is very similar to (S7.2). In this case, the map α from A[X] ⊗A M to M [X] takes f (X) ⊗ m to f (X)m, and the map β from M [X] to A[X] ⊗A M takes X i m to X i ⊗ m. Here, there is no need to show that β is well-deﬁned. ♣

page 1 of Solutions Chapters 1-5

SOLUTIONS CHAPTERS 1-5 Section 1.1 1. Under multiplication, the positive integers form a monoid but not a group, and the positive even integers form a semigroup but not a monoid. 2. With |a| denoting the order of a, we have |0| = 1, |1| = 6, |2| = 3, |3| = 2, |4| = 3, and |5| = 6. 3. There is a subgroup of order 6/d for each divisor d of 6. We have Z6 itself (d=1), {0}(d = 6), {0, 2, 4}(d = 2), and {0, 3}(d = 3). 4. S forms a group under addition. The inverse operation is subtraction, and the zero matrix is the additive identity. 5. S ∗ does not form a group under multiplication, since a nonzero matrix whose determinant is 0 does not have a multiplicative inverse. 6. If d is the smallest positive integer in H, then H consists of all multiples of d. For if x ∈ H we have x = qd + r where 0 ≤ r < d. But then r = x − qd ∈ H, so r must be 0. 7. Consider the rationals with addition mod 1, in other words identify rational numbers that diﬀer by an integer. Thus, for example, 1/3 = 4/3 = 7/3, etc. The group is inﬁnite, but every element generates a ﬁnite subgroup. For example, the subgroup generated by 1/3 is {1/3, 2/3, 0}. 8. (ab)mn = (am )n (bn )m = 1, so the order of ab divides mn. Thus |ab| = m1 n1 where m1 divides m and n1 divides n. Consequently, am1 n1 bm1 n1 = 1

(1)

If m = m1 m2 , raise both sides of (1) to the power m2 to get bmn1 = 1. The order of b, namely n, must divide mn1 , and since m and n are relatively prime, n must divide n1 . But n1 divides n, hence n = n1 . Similarly, if n = n1 n2 we raise both sides of (1) to the power n2 and conclude as above that m = m1 . But then |ab| = m1 n1 = mn, as asserted. If c belongs to both < a > and then since c is a power of a and also a power of b, we have cm = cn = 1. But then the order of c divides both m and n, and since m and n are relatively prime, c has order 1, i.e., c = 1. 9. Let |a| = m, |b| = n. If [m, n] is the least common multiple, and (m, n) the greatest common divisor, of m and n, then [m, n] = mn/(m, n). Examine the prime factorizations of m and n: t ti+1 m = (pt11 · · · ptii )(pi+1 · · · pjj ) = r r i+1 n = (pu1 1 · · · pui i )(pi+1 · · · pj j ) = s s

u

u

where tk ≤ uk for 1 ≤ k ≤ i, and tk ≥ uk for i + 1 ≤ k ≤ j. Now ar has order m/r and bs has order n/s, with m/r (= r ) and n/s (= s ) relatively prime. By Problem 8, ar bs has order mn/rs = mn/(m, n) = [m, n]. Thus given elements of orders m and n, we can construct another element whose order is the least common multiple of m and n. Since the least common multiple of m, n and q is [[m, n], q], we can inductively ﬁnd an element whose order is the least common multiple of the orders of all elements of G. 10. Choose an element a that belongs to H but not K, and an element b that belongs to K but not H, where H and K are subgroups whose union is G. Then ab must belong to either H or K, say ab = h ∈ H. But then b = a−1 h ∈ H, a contradiction. If ab = k ∈ K, then a = kb−1 ∈ K, again a contradiction. To prove the last statement, note that if H ∪ K is a subgroup, the ﬁrst result with G replaced by H ∪ K implies that H = H ∪ K or K = H ∪ K, in other words, K ⊆ H or H ⊆ K. 11. akm = 1 if and only if km is a multiple of n, and the smallest such multiple occurs when km is the least common multiple of n and k. Thus the order of ak is [n, k]/k. Examination of the prime factorizations of n and k shows that [n, k]/k = n/(n, k).

page 2 of Solutions Chapters 1-5

12. We have x ∈ Ai iﬀ x is a multiple of pi , and there are exactly n/pi multiples of pi between 1 and n. Similarly, x belongs to Ai ∩ Aj iﬀ x is divisible by pi pj , and there are exactly pinpj multiples of pi pj between 1 and n. The same technique works for all other terms. 13. The set of positive integers in {1, 2, ..., n} and not relatively prime to n is ∪ri=1 Ai , so ϕ(n) = n − | ∪ri=1 Ai |. By the principle of inclusion and exclusion from basic combinatorics, |

r i=1

Ai | =

r

|Ai | −

i=1

|Ai ∩ Aj | +

i<j

|Ai ∩ Aj ∩ Ak | − · · · + (−1)r−1 |A1 ∩ A2 ∩ · · · Ar |.

i<j
By Problem 12, ϕ(n) = n[1 −

r 1 1 1 1 + − + · · · + (−1)r ]. p p p p p p p p · · · pr i i j i j k 1 2 i=1 i<j i<j
Thus ϕ(n) = n(1 −

1 p1 )(1

−

1 p2 ) · · · (1

−

1 pr ).

14. Let G be cyclic of prime order p. Since the only positive divisors of p are 1 and p, the only subgroups of G are G and {1}. 15. No. Any non-identity element of G generates a cyclic subgroup H. If H ⊂ G, we are ﬁnished. If H = G, then G is isomorphic to the integers, and therefore has many nontrivial proper subgroups. (See (1.1.4) and Problem 6 above.) Section 1.2 1. The cycle decomposition is (1,4)(2,6,5); there is one cycle of even length, so the permutation is odd. 2. The elements are I, R = (A, B, C, D), R2 = (A, C)(B, D), R3 = (A, D, C, B), F = (B, D), RF = (A, B)(C, D), R2 F = (A, C), R3 F = (A, D)(B, C). 3. Such a permutation can be written as (1, a1 , a2 , a3 , a4 ) where (a1 , a2 , a3 , a4 ) is a permutation of {2, 3, 4, 5}. Thus the number of permutations is 4! = 24. 4. Select two symbols from 5, then two symbols from the remaining 3, and divide by 2 since, for example, (1,4)(3,5) is the same as (3,5)(1,4). The number of permutations is 10(3)/2 = 15. 5. For example, (1,2,3)(1,2) = (1,3) but (1,2)(1,2,3) = (2,3). 6. We have V = {I, (1, 2)(3, 4), (1, 3)(2, 4), (1, 4)(2, 3)}. Thus V = {I, a, b, c} where the product of any two distinct elements from {a, b, c} (in either order) is the third element, and the square of each element is I. It follows that V is an abelian group. 7. This follows because the inverse of the cycle (a1 , a2 , . . . , ak ) is (ak , . . . , a2 , a1 ). 8. Pick 3 symbols out of 4 to be moved, then pick one of two possible orientations, e.g., (1,2,3) or (1,3,2). The number of 3-cycles in S4 is therefore 4(2) = 8. 9. If π is a 3-cycle, then π 3 = I, so π 4 = π. But π 4 = (π 2 )2 , and π 2 ∈ H by hypothesis, so (π 2 )2 ∈ H because H is a group. Thus π ∈ H. 10. There are 5 inversions, 21, 41, 51, 43 and 53. Thus we have an odd number of inversions and the permutation π = (1, 2, 4)(3, 5) is also odd. 11. This follows because a transposition of two adjacent symbols in the second row changes the number of inversions by exactly 1. Therefore such a transposition changes the parity of the number of inversions. Thus the parity of π coincides with the parity of the number of inversions. In the given example, it takes 5 transpositions of adjacent digits to bring 24513 into natural order 12345. It also takes 5 transpositions to create π: π = (1, 5)(1, 4)(1, 2)(3, 5)(3, 4)

page 3 of Solutions Chapters 1-5

Section 1.3 1. If Ha = Hb then a = 1a = hb for some h ∈ H, so ab−1 = h ∈ H. Conversely, if ab−1 = h ∈ H then Ha = Hhb = Hb. 2. Reﬂexivity: aa−1 = 1 ∈ H. Symmetry: If ab−1 ∈ H then (ab−1 )−1 = ba−1 ∈ H. Transitivity: If ab−1 ∈ H and bc−1 ∈ H then (ab−1 )(bc−1 ) = ac−1 ∈ H. 3. ab−1 ∈ H iﬀ (ab−1 )−1 = ba−1 ∈ H iﬀ b ∈ Ha. 4. Ha−1 = Hb−1 iﬀ a−1 (b−1 )−1 = a−1 b ∈ H iﬀ aH = bH. 5. Since a1 belongs to both aH and a1 H, we have a1 H = aH because the left cosets partition G. 6. There are only two left cosets of H in G; one is H itself, and the other is, say, aH. Similarly, there are only two right cosets, H and Hb. Since the left cosets partition G, as do the right cosets, aH must coincide with Hb, so that every left coset if a right coset. 7. The permutations on the list are e, (1, 2, 3), (1, 3, 2), (1, 2), (1, 3), and (2, 3), which are in fact the 6 distinct permutations of {1, 2, 3}. 8. The left cosets of H are H = {e, b}, aH = {a, ab}, and a2 H = {a2 , a2 b}. The right cosets of H are H = {e, b}, Ha = {a, ba} = {a, a2 b}, and Ha2 = {a2 , ba2 } = {a2 , ab}. 9. The computation of Problem 8 shows that the left cosets of H do not coincide with the right cosets. Explicitly, aH and a2 H are not right cosets (and similarly, Ha and Ha2 are not left cosets). 10. f (n) = f (1 + 1 + · · · 1) = f (1) + f (1) + · · · f (1) = r + r + · · · r = rn. 11. In Problem 10, the image f (Z) must coincide with Z. But f (Z) consists of all multiples of r, and the only way f (Z) can equal Z is for r to be ±1. 12. The automorphism group of Z is {I, −I} where (−I)2 = I. Thus the automorphisms of Z form a cyclic group of order 2. (There is only one such group, up to isomorphism.) 13. Reﬂexivity: x = 1x1. Symmetry: If x = hyk, then y = h−1 xk −1 . Transitivity: if x = h1 yk1 and y = h2 zk2 , then x = h1 h2 zk2 k1 . 14. HxK is the union over all k ∈ K of the right cosets H(xk), and also the union over all h ∈ H of the left cosets (hx)K. Section 1.4 1. Deﬁne f : Z → Zn by f (x) = the residue class of x mod n. Then f is an epimorphism with kernel nZ, and the result follows from the ﬁrst isomorphism theorem. 2. Deﬁne f : Zn → Zn/m by f (x) = x mod n/m. Then f is an epimorphism with kernel Zm , and the result follows from the ﬁrst isomorphism theorem. (In the concrete example with n = 12, m = 4, we have f (0) = 0, f (1) = 1, f (2) = 2, f (3) = 0, f (4) = 1, f (5) = 2, f (6) = 0, etc.) 3. f (xy) = axya−1 = axa−1 aya−1 = f (x)f (y), so f is a homomorphism. If b ∈ G, we can solve axa−1 = b for x, namely x = a−1 ba, so f is surjective. If axa−1 = 1 then ax = a, so x = 1 and f is injective. Thus f is an automorphism. 4. Note that fab (x) = abx(ab)−1 = a(bxb−1 )a−1 = fa (fb (x)), and y = fa (x) iﬀ x = fa−1 (y), so that (fa )−1 = fa−1 . 5. Deﬁne Ψ : G → Inn G, the group of inner automorphisms of G, by Ψ(a) = fa . Then Ψ(ab) = fab = fa ◦ fb = Ψ(a)Ψ(b), so Ψ is a homomorphism (see the solution to Problem 4). Since a is arbitrary, Ψ is surjective. Now a belongs to ker Ψ iﬀ fa is the identity function, i.e., axa−1 = x for all x ∈ G, in other words, a commutes with every x in G. Thus ker Ψ = Z(G), and the result follows from the ﬁrst isomorphism theorem. 6. If f is an automorphism of Zn , then since 1 generates Zn , f is completely determined by m = f (1), and since 1 has order n in Zn , m must have order n as well. But then m is a unit mod n (see (1.1.5)), and f (r) = f (1 + 1 + · · · 1) = f (1) + f (1) + · · · f (1) = rf (1) = rm. Conversely, any unit m mod n determines an automorphism θ(m) = multiplication by m. The correspondence between m and θ(m) is a group isomorphism because θ(m1 m2 ) = θ(m1 ) ◦ θ(m2 ). 7. The ﬁrst assertion follows from the observation that HN is the subgroup generated by

page 4 of Solutions Chapters 1-5

H in 8. 9.

∪ N (see (1.3.6)). For the second assertion, note that if K is a subgroup of G contained both H and N , then K is contained in H ∩ N . If g(x) = y, then g ◦ fa ◦ g −1 maps y to g(axa−1 ) = g(a)y[g(a)]−1 . If G is abelian, then fa (x) = axa−1 = aa−1 x = x.

Section 1.5 1. C2 × C2 has 4 elements 1 = (1, 1), α = (a, 1), β = (1, a) and γ = (a, a), and the product of any two distinct elements from {α, β, γ} is the third. Since each of α, β, γ has order 2 (and 1 has order 1), there is no element of order 4 and C2 × C2 is not cyclic. 2. The four group is V = {I, a, b, c} where the product of any two distinct elements from {a, b, c} is the third. Therefore, the correspondence 1 → I, α → a, β → b, γ → c is an isomorphism of C2 × C2 and V . 3. Let C2 = {1, a} with a2 = 1, and C3 = {1, b, b2 } with b3 = 1. Then (a, b) generates C2 × C3 , since the successive powers of this element are (a, b), (1, b2 ), (a, 1), (1, b), (a, b2 ), and (1,1). Therefore C2 × C3 is cyclic of order 6, i.e., isomorphic to C6 . 4. Proceed as in Problem 3. If a has order n in Cn and b has order m in Cm , then (a, b) has order nm in Cn × Cm , so that Cn × Cm is cyclic of order nm. 5. Suppose that (a, b) is a generator of the cyclic group Cn × Cm . Then a must generate Cn and b must generate Cm (recall that Cn × {1} can be identiﬁed with Cn ). But (a, b)k = 1 iﬀ ak = bk = 1, and it follows that the order of (a, b) is the least common multiple of the orders of a and b, i.e., the least common multiple of n and m. Since n and m are not relatively prime, the least common multiple is strictly smaller than nm, so that (a, b) cannot possibly generate Cn × Cm , a contradiction. 6. By (1.3.3), G and H are both cyclic. Since p and q are distinct primes, they are relatively prime, and by Problem 4, G × H is cyclic. 7. Deﬁne f : H × K → K × H by f (h, k) = (k, h). It follows from the deﬁnition of direct product that f is an isomorphism. 8. Deﬁne f1 : G × H × K → G × (H × K) by f1 (g, h, k) = (g, (h, k)), and deﬁne f2 : G × H × K → (G × H) × K by f2 (g, h, k) = ((g, h), k). It follows from the deﬁnition of direct product that f1 and f2 are isomorphisms. Section 2.1 1. Never. If f is a polynomial whose degree is at least 1, then f cannot have an inverse. For if f (X)g(X) = 1, then the leading coeﬃcient of g would have to be 0, which is impossible. 2. If f (X)g(X) = 1, then (see Problem 1) f and g are polynomials of degree 0, in other words, elements of R. Thus the units of R[X] are simply the nonzero elements of R. 3. (a) No element of the form a1 X + a2 X 2 + · · · can have an inverse. (b) For example, 1 − X is a unit because (1 − X)(1 + X + X 2 + X 3 + · · ·) = 1. 4. Since Z[i] is a subset of the ﬁeld C of complex numbers, there can be no zero divisors in Z[i]. If w is a nonzero Gaussian integer, then w has an inverse in C, but the inverse need not belong to Z[i]. For example, (1 + i)−1 = 12 − 12 i. 5. If z = a + bi with a and b integers, then |z|2 = a2 + b2 , so that if z is not zero, we must have |z| ≥ 1. Thus if zw = 1, so that |z||w| = 1, we have |z| = 1, and the only possibilities are a = 0, b = ±1 or a = ±1, b = 0. Consequently, the units of Z[i] are 1, −1, i and −i. 6. All identities follow directly from the deﬁnition of multiplication of quaternions. Alternatively, (b) can be deduced from (a) by interchanging x1 and x2 , y1 and y2 , z1 and z2 , and w1 and w2 . Then the second identity of (c) can be deduced by noting that the quaternion on the right side of the equals sign in (a) is the conjugate of the quaternion on the right side of the equals sign in (b). 7. Multiply identities (a) and (b), and use (c). (This is not how Euler discovered the identity; quaternions were not invented until much later.) 8. The veriﬁcation that End G is an abelian group under addition uses the fact that G is an abelian group. The additive identity is the zero function, and the additive inverse of f is given by (−f )(a) = −f (a). Multiplication is associative because composition of functions

page 5 of Solutions Chapters 1-5

is associative. To establish the distributive laws, note that the value of (f + g)h at the element a ∈ G is f (h(a)) + g(h(a)), so that (f + g)h = f h + gh. Furthermore, the value of f (g + h) at a is f (g(a) + h(a)) = f (g(a)) + f (h(a)) since f is an endomorphism. Therefore f (g + h) = f h + gh. The multiplicative identity is the identity function, given by E(a) = a for all a. 9. An endomorphism that has an inverse must be an isomorphism of G with itself. Thus the units of the ring End G are the automorphisms of G. 10. Use Euler’s identity with x1 = 1, y1 = 2, z1 = 2, w1 = 5 (34 = 12 + 22 + 22 + 52 ) and x2 = 1, y2 = 1, z2 = 4, w2 = 6 (54 = 12 + 12 + 42 + 62 ). The result is 1836 = (34)(54) = 412 + 92 + 52 + 72 . The decomposition is not unique; another possibility is x1 = 0, y1 = 0, z1 = 3, w1 = 5, x2 = 0, y2 = 1, z2 = 2, w2 = 7. 11. In all four cases, sums and products of matrices of the given type are also of that type. But in (b), there is no matrix of the given form that can serve as the identity.. Thus the sets (a), (c) and (d) are rings, but (b) is not. Section 2.2 1. By Section 1.1, Problem 6, the additive subgroups of Z are of the form (n) = all multiples of n. But if x ∈ (n) and r ∈ Z then rx ∈ (n), so each (n) is an ideal as well. 2. If the n by n matrix A is 0 except perhaps in column k, and B is any n by n matrix, then BA is 0 except perhaps in column k. Similarly, if A is 0 oﬀ row k, then so is AB. 3. (a) This follows from the deﬁnition of matrix multiplication. (b) In (a) we have ajr = 0 for r = k, and the result follows. (c) By (b), the ith term of the sum is a matrix with cik in the ik position, and 0’s elsewhere. The sum therefore coincides with C. 4. The statement about left ideals follows from the formula of Problem 3(c). The result for right ideals is proved in a similar fashion. Explicitly, AEij has column i of A as its j th column, with 0’s elsewhere. If A ∈ Rk then AEij has aki in the kj position, with 0’s elsewhere, so if aki = 0 we have AEij a−1 ki = Ekj . Thus if C ∈ Rk then n

AEij a−1 ki ckj = C.

j=1

5. If I is a two-sided ideal and A ∈ I with ars = 0, then by considering products of the form a−1 rs Epq AEkl (which have the eﬀect of selecting an entry of A and sliding it from one row or column to another), we can show that every matrix Eij belongs to I. Since every matrix is a linear combination of the Eij , it follows that I = Mn (R). 6. A polynomial with no constant term is of the form a1 X + a2 X 2 + · · · an X n = Xg(X). Conversely, a polynomial expressible as Xg(X) has no constant term. Thus we may take f = X. 7. Let a be a nonzero element of R. Then the principal ideal (a) is not {0}, so (a) = R. Thus 1 ∈ (a), so there is an element b ∈ R such that ab = 1. 8. Since an ideal I is a ﬁnite set in this case, it must have a ﬁnite set of generators x1 , . . . , xk . Let d be the greatest common divisor of the xi . Every element of I is of the form a1 x1 + · · · + ak xk , and hence is a multiple of d. Thus I ⊆ (d). But d ∈ I, because there are integers ai such that i ai xi = d. Consequently, (d) ⊆ I. [Technically, arithmetic is modulo n, but we get around this diﬃculty by noting that if ab = c as integers, then ab ≡ c modulo n.] Section 2.3 1. Use the same maps as before, and apply the ﬁrst isomorphism theorem for rings. 2. If In is the set of multiples of n > 1 in the ring of integers, then In is an ideal but not a subring (since 1 ∈ / In ). Z is a subring of the rational numbers Q but not an ideal, since a rational number times an integer need not be an integer.

page 6 of Solutions Chapters 1-5

3. In parts (2) and (3) of the Chinese remainder theorem, take R = Z and Ii = the set of multiples of mi . 4. Apply part (4) of the Chinese remainder theorem with R = Z and Ii = the set of multiples of mi . 5. To prove the ﬁrst statement, deﬁne f : R → R2 by f (r1 , r2 ) = r2 . Then f is a ring homomorphism with kernel R1 and image R2 . By the ﬁrst isomorphism theorem for rings, R/R1 ∼ = R2 . A symmetrical argument proves the second statement. In practice, we tend to forget about the primes and write R/R1 ∼ = R2 and R/R2 ∼ = R1 . There is also a tendency to identify a ring with its isomorphic copy, and write R/R1 = R2 and R/R2 = R1 This should not cause any diﬃculty if you add, mentally at least, ”up to isomorphism”. 6. The product is always a subset of the intersection, by deﬁnition. Conversely, if a ∈ ∩i Ii , then (a + I1 , . . . , a + In ) is zero in the direct product. By part (4) of the Chinese remainder theorem, a + ∩i Ii is zero in R/ ∩i Ii . Thus a belongs to the intersection of the Ii . 7. Let a + ∩i Ii map to (1 + I1 , 0 + I2 , c3 + I3 , . . . , cn + In ), where the cj are arbitrary. Then 1 − a ∈ I1 and a ∈ I2 , so 1 = (1 − a) + a ∈ I1 + I2 . Thus I1 + I2 = R, and similarly Ii + Ij = R for all i = j. Section 2.4 1. If n = rs with r, s > 1 then r ∈< / n >, s ∈< / n >, but rs ∈< n >, so that < n > is not prime. But Z/ < n > is isomorphic to Zn , the ring of integers modulo n (see Section 2.3, Problem 1). If n is prime, then Zn is a ﬁeld, in particular an integral domain, hence < n > is a prime ideal by (2.4.5). 2. By Problem 1, I is of the form where p is prime. Since Z/ is isomorphic to Zp , which is a ﬁeld, is maximal by (2.4.3). 3. The epimorphism a0 + a1 X + a2 X 2 + · · · → a0 of F [[X]] onto F has kernel < X >, and the result follows from (2.4.7). 4. The ideal I =< 2, X > is not proper; since 2 ∈ I and 12 ∈ F ⊆ F [[X]], we have 1 ∈ I and therefore I = F [[X]]. The key point is that F is a ﬁeld, whereas Z is not. 5. Suppose that f (X) = a0 + a1 X + · · · belongs to I but not to < X >. Then a0 cannot be 0, so by ordinary long division we can ﬁnd g(X) ∈ F [[X]] such that f (X)g(X) = 1. But then 1 ∈ I, contradicting the assumption that I is proper. 6. Let f (X) = an X n + an+1 X n+1 + · · · , an = 0, be an element of the ideal I, with n as small as possible. Then f (X) ∈ (X n ), and if g(X) is any element of I, we have g(X) ∈ (X m ) for some m ≥ n. Thus I ⊆ (X n ). Conversely, if f (X) = X n g(X) ∈ I, with g(X) = an + an+1 X + · · · , an = 0,, then as in Problem 5, g(X) is a unit, and therefore X n ∈ I. Thus (X n ) ⊆ I, so that I = (X n ), as claimed. 7. f −1 (P ) is an additive subgroup by (1.3.15), part (ii). If a ∈ f −1 (P ) and r ∈ R, then f (ra) = f (r)f (a) ∈ P , so ra ∈ f −1 (P ). Thus f −1 (P ) is an ideal. If ab ∈ f −1 (P ), then f (a)f (b) ∈ P , so either a or b must belong to f −1 (P ). Consequently, f −1 (P ) is prime. 8. Let S be a ﬁeld, and R an integral domain contained in S, and assume that R is not a ﬁeld. For example, let R = Z, S = Q. Take f to be the inclusion map. Then {0} is a maximal ideal of S, but f −1 ({0}) = {0} is a prime but not maximal ideal of R. 9. If P = I ∩ J with P ⊂ I and P ⊂ J, choose a ∈ I \ P and b ∈ J \ P . Then ab ∈ I ∩ J = P , contradicting the assumption that P is prime. Section 2.5 1. Any number that divides a and b divides b and r1 , and conversely, any number that divides b and r1 divides a and b. Iterating this argument, we ﬁnd that gcd(a, b) = gcd(b, r1 ) = gcd(r1 , r2 ) = · · · = gcd(rj−1 , rj ) = rj . 2. This follows by successive substitution. We start with rj = rj−2 − rj−1 qj , continue with rj−1 = rj−3 − rj−2 qj−1 , rj−2 = rj−4 − rj−3 qj−2 , and proceed up the ladder until we have expressed d as a linear combination of a and b. There is an easier way, as Problem 3 shows. 3. The ﬁrst equation of the three describes the steps of the algorithm. We wish to prove

page 7 of Solutions Chapters 1-5

that axi + byi = ri , that is, a(xi−2 − qi xi−1 ) + b(yi−2 − qi yi−1 ) = ri .

(1)

But this follows by induction: if axi−2 + byi−2 = ri−2 and axi−1 + byi−1 = ri−1 , then the left side of (1) is ri−2 − qi ri−1 , which is ri by deﬁnition of the Euclidean algorithm. 4. We have the following table: i -1 0 1 2 3 4

qi+1 — 2 3 1 1 2

ri 123 54 15 9 6 3

xi 1 0 1 -3 4 -7

yi 0 1 -2 7 -9 16

For example, to go from i = 1 to i = 2 we have x2 = x0 − q2 x1 = 0 − 3(1) = −3, y2 = y0 − q2 y1 = 1 − 3(−2) = 7, and r2 = r0 − q2 r1 = 54 − 3(15) = 9; also, q3 = 15/9 = 1. We have ax2 + by2 = 123(−3) + 54(7) = 9 = r2 , as expected. The process terminates with 123 (-7) + 54 (16) = 3 = d. 5. If p is composite, say p = rs with 1 < r < p, 1 < s < p, then rs is 0 in Zp but r and s are nonzero, so Zp is not a ﬁeld. If p is prime and a is not zero in Zp then the greatest common divisor of a and p is 1, and consequently there are integers x and y such that ax + py = 1. In Zp this becomes ax = 1, so that every nonzero element in Zp has an inverse in Zp , proving that Zp is a ﬁeld. 6. Since f (X) and g(X) are multiples of d(X), so are all linear combinations a(X)f (X) + b(X)g(X), and consequently I ⊆ J. By Problem 2, there are polynomials a(X) and b(X) such that a(X)f (X) + b(X)g(X) = d(X), so that d(X) belongs to I. Since I is an ideal, every multiple of d(X) belongs to I, and therefore J ⊆ I. n 7. Take f (X) = i=0 bi Pi (X). 8. If g(X) is another polynomial such that g(ai ) = f (ai ) for all i, then f and g agree at n + 1 points, so that f (X) − g(X) has more than n roots in F . By (2.5.3), f (X) − g(X) must be the zero polynomial. 9. If F has only ﬁnitely many elements a1 , . . . , an , take f (X) = (X − a1 ) · · · (X − an ). 10. Let F be the complex numbers C. Then every polynomial of degree n has exactly n roots, counting multiplicity. Thus if f (a) = 0 at more than n points a, in particular if f vanishes at every point of C, then f = 0. M:ore generally, F can be any inﬁnite ﬁeld (use (2.5.3)). Section 2.6 1. If r = 0 then I contains a unit, so that 1 ∈ I and I = R. 2. If b ∈< / p1 > then b+ < p1 >=< p1 >, so b+ < p1 > has an inverse in R/ < p1 >, say c+ < p1 >. Thus (b+ < p1 >)(c+ < p1 >) = 1+ < p1 >, hence (bc − 1)+ < p1 >=< p1 >, so bc − 1 ∈< p1 >. 3. If bc − dp1 = 1 then bcp2 · · · pn − dp1 · · · pn = p2 · · · pn , and since b and p1 · · · pn belong to I, so does p2 · · · pn , contradicting the minimality of n. (If n = 1, then 1 ∈ I, so I = R.) 4. If a, b ∈ R and x, y ∈ J then (ax + by)p1 = xp1 a + yp1 b. Since x, y ∈ J we have xp1 ∈ I and yp1 ∈ I, so that (ax + by)p1 ∈ I, hence ax + by ∈ J. 5. If x ∈ J then xp1 ∈ I, so Jp1 ⊆ I. Now I ⊆< p1 > by Problem 3, so if a ∈ I then a = xp1 for some x ∈ R. But then x ∈ J by deﬁnition of J, so a = xp1 ∈ Jp1 . 6. Since J contains a product of fewer than n primes, J is principal by the induction hypothesis. If J =< d > then by Problem 5, I = J < p1 >. But then I =< dp1 >, and the result follows. (If n = 1, then p1 ∈ I, hence 1 ∈ J , so J = R and I = J < p1 >=< p1 >.) 7. Assume that P ⊆ Q. Then p = aq for some a ∈ R, so aq ∈ P . Since P is prime, either

page 8 of Solutions Chapters 1-5

a or q belongs to P . In the second case, Q ⊆ P and we are ﬁnished. Thus assume a ∈ P , so that a = bp for some b ∈ R. Then p = aq = bpq, and since R is an integral domain and p = 0, we have bq = 1, so q is a unit and Q = R, a contradiction of the assumption that Q is prime. 8. Let x be a nonzero element of P , with x = up1 · · · pn , u a unit and the pi prime. Then p1 · · · pn = u−1 x ∈ P , and since P is prime, some pi belongs to P . Thus P contains the nonzero principal prime ideal < pi >. Section 2.7 1. If m is a generator of the indicated ideal then m belongs to all < ai >, so each ai divides m. If each ai divides b then b is in every < ai >, so b ∈ ∩ni=1 < ai >=< m >, so m divides b. Thus m is a least common multiple of A. Now suppose that m is an lcm of A, and let ∩ni=1 < ai >=< c >. Then c belongs to every < ai >, so each ai divides c. Since m =lcm(A), m divides c, so < c > is a subset of < m >. But again since m = lcm(A), each ai divides m, so m ∈ ∩ni=1 < ai >=< c >. Therefore < m >⊆< c >, hence < m >=< c >, and m is a generator of ∩ni=1 < ai >. 2. Let a = 11 + 3i, b = 8 − i. Then a/b = (11 + 3i)(8 + i)/65 = 85/65 + i35/65. Thus we may take x0 = y0 = 1, and the ﬁrst quotient is q1 = 1 + i. The ﬁrst remainder is r1 = a − bq1 = (11 + 3i) − (8 − i)(1 + i) = 2 − 4i. The next step in the Euclidean algorithm is (8 − i)/(2 − 4i) = (8 − i)(2 + 4i)/20 = 1 + (3i/2). Thus the second quotient is q2 = 1 + i (q2 = 1 + 2i would be equally good). The second remainder is r2 = (8 − i) − (2 − 4i)(1 + i) = 2 + i. The next step is (2 − 4i)/(2 + i) = (2 − 4i)(2 − i)/5 = −2i, so q3 = −2i, r3 = 0. The gcd is the last divisor, namely 2 + i. 3. We have Ψ(1) ≤ Ψ(1(a)) = Ψ(a) for every nonzero a. If a is a unit with ab = 1, then Ψ(a) ≤ Ψ(ab) = Ψ(1), so Ψ(a) = Ψ(1). Conversely, suppose that a = 0 and Ψ(a) = Ψ(1). Divide 1 by a to get 1 = aq + r, where r = 0 or Ψ(r) < Ψ(a) = Ψ(1). But if r = 0 then Ψ(r) must be greater than or equal to Ψ(1), so we must have r = 0. Therefore 1 = aq, and a is a unit. √ √ √ 4. Ψ((a1 + b1 d)(a2 + b2 d)) = ψ(a1 a2 + b1 b2 d + (a1 b2 + a2 b1 ) d) √ √ = |a1 a2 + b1 b2 d + (a1 b2 + a2 b1 ) d||a1 a2 + b1 b2 d − (a1 b2 + a2 b1 ) d|; √ √ √ √ √ √ Ψ(a1 + b1 d)Ψ(a2 + b2 d) = |a1 + b1 d||a2 + b2 d||a1 − b1 d||a2 − b2 d| and it follows that Ψ(αβ) = Ψ(α)Ψ(β). Now Ψ(α) ≥ 1 for all nonzero α, for if Ψ(α) = |a2 − db2 | = 0, then a2 = db2 . But if b = 0 then d = (a/b)2 , contradicting the assumption that d is not a perfect square. Thus b = 0, so a is 0 as well, and α = 0, a contradiction. Thus Ψ(αβ) = Ψ(α)Ψ(β) ≥ Ψ(α). √ √ 2 5. Either d or d√− 1 is even, − d = (d + √ d)(d − d). But 2 does √ so 2 divides d(d − 1) = d √ not divide d + √ d or d − d. For example, if √ 2(a + b d) = d + √d) for integers a, b then 2a − d = (1 − 2b) d, which is impossible since d is irrational. (If d = r/s then r2 = ds2 , which cannot happen if d is not a perfect square.) 6. Deﬁne Ψ as√in Problem 4 (and Example (2.7.5)). Suppose 2 = αβ where α and β are nonunits in Z[ d]. Then 4 = Ψ(2) = Ψ(α)Ψ(β), with Ψ(α), Ψ(β) > 1 by Problems 3 and 4. √ But then Ψ(α) = Ψ(β) = 2. If α = a + b d then |a2 − db2 | = 2, so a2 − db2 is either 2 or −2. Therefore if b = 0 (so that b2 ≥ 1), then since d ≤ −3 we have a2 − db2 ≥ 0 + 3(1) = 3, a contradiction. Thus b = 0, so α = a, and 2 = Ψ(a) = a2 , an impossibility for a ∈ Z. 7. This follows from Problems 5 and 6, along with (2.6.4). 8. Just as with ordinary integers, the product of two Gaussian integers is their greatest common divisor times their least common multiple. Thus by Problem 2, the lcm is (11 + 3i)(8 − i)/(2 + i) = 39 − 13i. 9. If α = βγ, then Ψ(α) = Ψ(β)Ψ(γ). By hypothesis, either Ψ(β) or Ψ(γ) is 1(= Ψ(1)). By Problem 3, either β or γ is a unit.

page 9 of Solutions Chapters 1-5

Section 2.8 1. If D is a ﬁeld, then the quotient ﬁeld F , which can be viewed as the smallest ﬁeld containing D, is D itself. Strictly speaking, F is isomorphic to D; the embedding map f (a) = a/1 is surjective, hence an isomorphism. To see this, note that if a/b ∈ F , then a/b = ab−1 /1 = f (ab−1 ). 2. The quotient ﬁeld consists of all rational functions f (X)/g(X), where f (X) and g(X) are polynomials in F [X] and g(X) is not the zero polynomial. To see this, note that the collection of rational functions is in fact a ﬁeld, and any ﬁeld containing F [X] must contain all such rational functions. +bde 3. ab + ( dc + fe ) and ( ab + dc ) + fe both compute to be adf +bcf . bdf 4.

a c b (d

+de + fe ) = ab ( cfdf )=

acf +ade bdf

and

+dae +ade ae + bf = acbfb2+bdae = acfbdf = acfbdf . df 5. If g is any extension of h and a/b ∈ F , there is only one possible choice for g(a/b), namely h(a)/h(b). (Since b = 0 and h is a monomorphism, h(b) = 0.) If we deﬁne g this way, then g(a) = g(a/1) = h(a)/h(1) = h(a), so g is in fact an extension of f . Furthermore, if a/b = c/d then since h is a monomorphism, h(a)/h(b) = h(c)/h(d). Therefore g is welldeﬁned. Again since h is a monomorphism, it follows that g( ab + dc ) = g( ab ) + g( dc ) and g( ab dc ) = g( ab )g( dc ). Since g is an extension of h, we have g(1) = 1, so g is a homomorphism. Finally, if g(a/b) = 0, then h(a) = 0, so a = 0 by injectivity of h. Thus g is a monomorphism. 6. The problem is that h is not injective. As before, if g is to be an extension of h, we must have g(a/b) = h(a)/h(b). But if b is a multiple of p, then h(b) is zero, so no such g can exist. 7. We must have g(a/b) = g(a/1)g((b/1)−1 ) = g(a)g(b)−1 . ac bd

8. If a/b = c/d, then for some s ∈ S we have s(ad−bc) = 0. So g(s)[g(a)g(d)−g(b)g(c)] = 0. Since g(s) is a unit, we may multiply by its inverse to get g(a)g(d) = g(b)g(c), hence g(a)g(b)−1 = g(c)g(d)−1 , proving that g is well-deﬁned. To show that g is a homomorphism, we compute c a ad + bc ) = g(ad + bc)g(bd)−1 g( + ) = g( b d bd a c = [g(a)g(d) + g(b)g(c)]g(b)−1 g(d)−1 = g( ) + g( ) b d Similarly, we have g( ab dc ) = g( ab )g( dc ) and g(1) = 1. Section 2.9 1. We have an (u/v)n + an−1 (u/v)n−1 + · · · + a1 (u/v) + a0 = 0; multiply by v n to get an un + an−1 un−1 v + · · · + a1 uv n−1 + a0 v n = 0. Therefore an un = −an−1 un−1 v − · · · − a1 uv n−1 − a0 v n . Since v divides the right side of this equation, it must divide the left side as well, and since u and v are relatively prime, v must divide an . Similarly, a0 v n = −an un − an−1 un−1 v − · · · − a1 uv n−1 , so u divides a0 . 2. X n − p satisﬁes Eisenstein’s criterion, and since the polynomial is primitive, it is irreducible over Z. 3. f3 (X) = X 3 + 2X + 1, which is irreducible over Z3 . For if f3 (X) were reducible over Z3 , it would have a linear factor (since it is a cubic), necessarily X − 1 or X + 1(= X − 2). But then 1 or 2 would be a root of f3 , a contradiction since f3 (1) = 1 and f3 (2) = 1 (mod 3). 4. By Eisenstein, X 4 + 3 is irreducible over Z. The substitution X = Y + 1 yields

page 10 of Solutions Chapters 1-5

Y 4 +4Y 3 +6Y 2 +4Y +4, which is therefore irreducible in Z[Y ]. Thus X 4 +4X 3 +6X 2 +4X +4 is irreducible in Z[X], i.e., irreducible over Z. 5. Note that < n, X > is a proper ideal since it cannot contain 1. If < n, X >=< f > then n ∈< f >, so n is a multiple of f . Thus f is constant (= 1), in which case X ∈< / f >. 6. Since 1 ∈< / X, Y >, < X, Y > is a proper ideal. Suppose < X, Y >=< f >. Then Y is a multiple of f , so f is a polynomial in Y alone (in fact f = cY ). But then X ∈< / f >, a contradiction. 7. If p = X + i, then p is irreducible since X + i is of degree 1. Furthermore, p divides X 2 + 1 but p2 does not. Take the ring R to be C[X, Y ] = (C[X])[Y ] and apply Eisenstein’s criterion. 8. Write f (X, Y ) as Y 3 + (X 3 + 1) and take p = X + 1. Since X 3 + 1 = (X + 1)(X 2 − X + 1) and X + 1 does not divide X 2 − X + 1, the result follows as in Problem 7. Section 3.1 1. F (S) consists of all quotients of ﬁnite linear combinations (with coeﬃcients in F ) of ﬁnite products of elements of S. To prove this, note ﬁrst that the set A of all such quotients is a ﬁeld. Then observe that any ﬁeld containing F and S must contain A, in particular, A ⊆ F (S). But F (S) is the smallest subﬁeld containing F and S, so F (S) ⊆ A. 2. The composite consists of all quotients of ﬁnite sums of products of the form xi1 xi2 · · · xin , n = 1, 2, . . ., where i1 , i2 , . . . , in ∈ I and xij ∈ Kij . As in Problem 1, the set A of all such quotients is a ﬁeld, and any ﬁeld that contains all the Ki must contain A. 3. By (3.1.9), [F [α] : F ] = [F [α] : F [β]][F [β] : F ], and since the degree of any extension is at least 1, the result √ follows. 2 4. Let min(−1 + 2, Q)=a0 + a√ 1 cannot work 1 X + X (a√polynomial of degree√ √ because √ 2 / Q). Then a0 +a1 (−1+ 2)+(−1+ 2)2 = 0. Since (−1+ 2) = 3−2 2, we have −1+ 2 ∈ √ a0 −a1 +3 = 0 and a1 −2 = 0, so a0 = −1, a1 = 2. Therefore min(−1+2 2, Q) = X 2 +2X−1. 5. Let β = b0 + b1 α + · · · + bn−1 αn−1 . Then for some a0 , . . . , an ∈ F we have a0 + a1 β + · · · + an β n = 0. Substituting the expression for β in terms of α into this equation, reducing to a polynomial in α of degree at most n − 1 (as in the proof of (3.1.7)), and setting the coeﬃcients of the αi , i = 0, 1, . . . , n − 1 equal to zero (remember that the αi form a basis for F [α] over F ), we get n linear equations in the n + 1 unknowns ai , i = 0, . . . , n. We know that a solution exists because β is algebraic over F . By brute force (try ai = 1, aj = 0, j > i for i = 1, 2, . . . , n), we will eventually arrive at the minimal polynomial. 6. Deﬁne ϕ : F (X) → E by ϕ(f (X)/g(X)) = f (α)/g(α). Note that ϕ is well-deﬁned, since if g is a nonzero polynomial, then g(α) = 0 (because α is transcendental over F ). By (3.1.2), ϕ is a monomorphism. Since ϕ(F (X)) = F (α), it follows that F (X) and F (α) are isomorphic. 7. The kernel of ϕ is I, and as in (3.1.3), F [X]/I is a ﬁeld. The image of ϕ is F [α], and by the ﬁrst isomorphism theorem for rings, F [α] is isomorphic to F [X]/I. Therefore F [α] is a ﬁeld, and consequently F [α] = F (α). 8. If f = gh, then (g + I)(h + I) = 0 in F [X]/I, so F [X]/I is not a ﬁeld. By (2.4.3), I is not maximal. 9. The minimal polynomial over F belongs to F [X] ⊆ E[X], and has α as a root. Thus min(α, E) divides min(α, F ). 10. The result is true for n = 1; see (3.1.7). Let E = F [α1 , . . . , αn−1 ], so that [F [α1 , . . . , αn ] : F ] = [F [α1 , . . . , αn ] : E][E : F ] = [E[αn ] : E][E : F ]. But [E[αn ] : E] is the degree of the minimal polynomial of αn over E, which is at most the degree of the minimal polynomial of αn over F , by Problem 9. An application of the induction hypothesis completes the proof. Section 3.2 1. f (X) = (X − 2)2 , so we may take the splitting √ ﬁeld K to be√Q itself. √ 2.√ f (X) = (X − 1)2 + 3, with roots 1√± i 3, so K = Q(i 3). Now i 3 ∈ / Q since (i 3)2 = −3 < 0, so [K : Q] ≥ 2. But i 3 is a root of X 2 + 3, so [K : Q] ≤ 2. Therefore [K : Q] = 2.

page 11 of Solutions Chapters 1-5

3. Let α be the positive 4th root of 2. The roots of f (X) are α, iα, −α and −iα. Thus K = Q(α, i). Now f (X) is irreducible by Eisenstein, so [Q(α) : Q] = 4. Since i ∈ / Q(α) and i is a root of X 2 + 1 ∈ Q(α)[X], we have [K : Q(α)] = 2. By (3.1.9), [K : Q] = 2 × 4 = 8. 4. The argument of (3.2.1) may be reproduced, with the polynomial f replaced by the family C of polynomials, and the roots α1 , . . . , αk of f by the collection of all roots of the polynomials in the family C. 5. Take f = f1 · · · fr . Since α is a root of f iﬀ α √ is a root of some fi , the√result follows. √ 6. If the degree is less than 4, it must be 2 (since m∈ / Q). In this case, n = a + b m, so √ n = a2 + b2 m + 2ab m. Since m is square-free, √ we must √ have a = 0 or b = 0, and the latter is impossible because n is square-free. Thus n = b m, so n = b2 m, again a contradiction of the hypothesis that n is square-free. Section 3.3 1. If α1 , . . . , αn form a basis for E over F , then E is generated over F by the αi . Each αi is algebraic over F because F (αi ) ⊆ E, and (3.1.10) applies. 2. There are only countably many polynomials with rational coeﬃcients, and each such polynomial has only ﬁnitely many roots. Since an algebraic number must be a root of one of these polynomials, the set of algebraic numbers is countable. Since the complex ﬁeld is uncountably inﬁnite, there are uncountably many transcendental numbers. 3. The complex ﬁeld C is algebraically closed, and C is an extension of the rational ﬁeld Q. But C is not algebraic over Q, by Problem 2. 4. The algebraic numbers A form a ﬁeld by (3.3.4), and A is algebraic over Q by deﬁnition. But it follows from Section 2.9, Problem 2, that A contains subﬁelds of arbitrarily high degree (in fact subﬁelds of every degree) over Q, so that A/Q is not ﬁnite. 5. This can be veriﬁed by transﬁnite induction. A splitting ﬁeld is always an algebraic extension (see (3.2.2)), and the ﬁeld F
page 12 of Solutions Chapters 1-5

F (α1p ), so each element of F (α1p , α2p ) can be written as a polynomial in α2p with coeﬃcients in F (α1p ). Such an element has the form ( brs α1pr )α2ps s

r

with the brs ∈ F . An induction argument completes the proof. 6. Extend the yi to a basis y1 , . . . , yn for E over F . By Problem 5, every element of E(= F (E p )) has the form y = a1 y1p + · · · + an ynp with the ai ∈ F . Thus {y1p , . . . , ynp } spans E over F . It follows that this set contains a basis, hence (since there are exactly n vectors in the set) the set is a basis for E over F . The result follows. 7. Assume the extension is separable, and let α ∈ E. Then α is separable over F , hence over F (αp ), so by Problem 3, α ∈ F (E p ). Thus E = F (E p ). Conversely, suppose that E = F (E p ) and the element α ∈ E has an inseparable minimal polynomial m(X). By (3.4.3), m(X) is of the form b0 +b1 X p +· · ·+br−1 X (r−1)p +X rp . Since m(α) = 0, the elements 1, αp , . . . , αrp are linearly dependent over F . But by minimality of m(X), 1, α, . . . , αrp−1 are linearly independent over F , hence 1, α, . . . , αr are linearly independent over F . (Note that rp − 1 ≥ 2r − 1 ≥ r.) By Problem 6, 1, αp , . . . , αrp are linearly independent over F , which is a contradiction. Thus E/F is separable. 8. We may assume that F has prime characteristic p. By Problem 7, E = K(E p ) and K = F (K p ). Thus E = F (K p , E p ) = F (E p ) since K ≤ E. Again by Problem 7, E/F is separable. m 9. If g can be factored, so can f , and therefore g is irreducible. If f (X) = g(X p ) with m maximal, then g ∈ / F [X p ]. By (3.4.3) part (2), g is separable. m 10. Suppose that the roots of g in a splitting ﬁeld are c1 , . . . , cr . Then f (X) = g(X p ) = m m (X p − c1 ) · · · (X p − cr ). By separability of g, the cj must be distinct, and since f (α) = 0, m m we have αp = cj for all j. This is impossible unless r = 1, in which case f (X) = X p − c1 . m But f ∈ F [X], so αp = c1 ∈ F . n n n 11. If αp = c ∈ F , then α is a root of X p − c = (X − α)p , so min(α, F ) is a power of X − α, and therefore has only one distinct root α. The converse follows from Problem 10 with f = min(α, F ). Section 3.5

√ 1. Take F = Q, K = Q( 3 2) (see (3.5.3)), and let E be any extension of K that is normal over F , for example, E = C. 2. The polynomial f (X) = X 2 − a is irreducible, else it would factor as (X − b)(X − c) with b + c = 0, bc = a, i.e. (X − b)(X + b) with b2 = a, contradicting the hypothesis. Thus E √ is obtained from Q by adjoining a root of f . The other root of f is − a, so that E is a splitting ﬁeld of f over Q.√By (3.5.7), E/Q √ is normal. 3. Take F = Q, K = Q( 2), E = Q( 4 2). Then K/F is normal by Problem 2, and E/K is normal by a similar argument. But E/F is not normal, since the two complex roots of X 4 − 2 do not belong to E. The same argument works with 2 replaced by any positive integer that is not a perfect square. 4. There are at most n embeddings of E in C extending σ. The proof is the same, except that now g has at most r distinct roots in C, so there are at most r possible choices of β. The induction hypothesis yields at most n/r extensions from F (α) to E, and the result follows. 5. Since the rationals have characteristic zero, the extension is separable. Since E is the 2 2 splitting ﬁeld √ of (X√ − 2)(X − 3) over Q, the extension is normal, hence Galois. 6. Since 3 ∈ / Q( 2), the extension has degree 4. By (3.5.9), there are exactly four Qautomorphisms in the Galois group. By (3.5.1), each such Q-automorphism must permute the roots of X 2 − 2 and must also permute the roots of X 2 − 3. There are only four possible √ ways√this can be done. Since a Q-automorphism is completely speciﬁed by its action on 2 and 3, the Galois group may be described as follows:

page 13 of Solutions Chapters 1-5

√ √ √ √ √ √ 2, 3 → 3; (2) 2 → 2, 3 → − 3; √ √ √ √ √ √ √ √ (3) 2 → − 2, 3 → 3; (4) 2 → − 2, 3 → − 3. Since the product (composition) of any two of automorphisms (2),(3),(4) is the third, the Galois group is isomorphic to the four group (Section 1.2, Problem 6). 7. Yes, up to isomorphism. If f is the polynomial given in (3.5.11), any normal closure is a splitting ﬁeld for f over F , and the result follows from (3.2.5). 8. If f is irreducible over F and has a root in E1 ∩ E2 , then f splits over both E1 and E2 , hence all roots of f lie in E1 ∩ E2 . Thus f splits over E1 ∩ E2 , and the result follows. (1)

√

2→

√

Section 4.1 1. If x ∈ R, take r(x + I) to be rx + I to produce a left R-module, and (x + I)r = xr + I for a right R-module. Since I is an ideal, the scalar multiplication is well-deﬁned, and the requirements for a module can be veriﬁed using the basic properties of quotient rings. 2. If A is an algebra over F , the map x → x1 of F into A is a homomorphism, and since F is a ﬁeld, it is a monomorphism (see (3.1.2)). Thus A contains a copy of F . Conversely, if F is a subring of A, then A is a vector space over F , and the compatibility conditions are automatic since A is commutative. 3. Let R = Z and let M be the additive group of integers mod m, where m is composite, say m = ab with a, b > 1. Take x = a (mod m) and r = b. 4. Any set containing 0 is linearly dependent, so assume a/b and c/d are nonzero rationals. Since a/b c/d is rational, the result follows. 5. In view of Problem 4, the only hope is that a single nonzero rational number a/b spans M over Z. But this cannot happen, since an integer multiple of a/b must be a fraction whose denominator is a divisor of b. 6. If a ∈ A ⊆ C and x ∈ B ∩ C, then ax ∈ (AB) ∩ C. Conversely, let c = ab ∈ (AB) ∩ C. Then b = a−1 c ∈ C since A ⊆ C. Thus ab ∈ A(B ∩ C). 7. If f (X) = a0 + a1 X + · · · + an X n and v ∈ V , take f (X)v = f (T )v = a0 Iv + a1 T v + · · · + an T n v where I is the identity transformation and T i is the composition of T with itself i times. Section 4.2 1. Let W be a submodule of M/N . By the correspondence theorem, W = L/N for some submodule L of M with L ≥ N . Since L = L + N , we have W = (L + N )/N . 2. No. If S is any submodule of M , then S + N is a submodule of M containing N , so S + N corresponds to W = (S + N )/N . We know that W can also be written as (L + N )/N where L ≥ N . (For example, L = S + N .) By the correspondence theorem, S + N = L + N , and there is no contradiction. 3. If A ∈ Mn (R), then AE11 retains column 1 of A, with all other columns zero. 4. To identify the annihilator of E11 , observe that by Problem 4, AE11 = 0 iﬀ column 1 of A is zero. For the annihilator of M , note that Ej1 ∈ M for every j, and AEj1 has column j of A as column 1, with zeros elsewhere. (See Section 2.2, Problem 4.) Thus if A annihilates everything in M , then column j of A is zero for every j, so that A is the zero matrix. 5. The quotient ring is an R-module by Problem 1 of Section 4.1 If r ∈ R then r+I = r(1+I), so R/I is cyclic with generator 1 + I. 6. We must show that scalar multiplication is well-deﬁned, that is, if r ∈ I, then rm = 0 for all m ∈ M . Thus I must annihilate M , in other words, IM = 0, where the submodule IM is the set of all ﬁnite sums rj mj , rj ∈ R, mj ∈ M . 7. No, since (r + I)m coincides with rm. Section 4.3 1. Essentially the same proof as in (4.3.3) works. If z1 + · · · + zn = 0, with zi ∈ Mi , then zn is a sum of terms from previous modules, and is therefore 0. Inductively, every zi is 0.

page 14 of Solutions Chapters 1-5

(In the terminology of (4.3.3), zi is xi − yi .) 2. Only when A = {0}. If A has n elements, then by Lagrange’s theorem, nx = 0 for every x ∈ A, so there are no linearly independent sets (except the empty set). 3. This follows because (−s)r + rs = 0. 4. If I is not a principal ideal, then I can never be free. For if I has a basis consisting of a single element, then I is principal, a contradiction. But by Problem 3, there cannot be a basis with more than one element. If I =< a > is principal, then I is free if and only if a is not a zero-divisor. 5. Z, or any direct sum of copies of Z, is a free Z-module. The additive group of rational numbers is not a free Z-module, by Problem 5 of Section 4.1 6. The “only if” part was done in (4.3.6), so assume that M has the given property. Construct a free module M = ⊕i∈S Ri where Ri = R for all i. Then the map f : S → M with f (i) = ei (where ei has 1 in its ith component and zeros elsewhere) extends to a homomorphism (also called f ) from M to M . Let g : M → M be the module homomorphism determined by g(ei ) = i. Then g ◦ f is the identity on S, hence on M , by the uniqueness assumption. Similarly, f ◦ g = 1. 7. An element of M is speciﬁed by choosing a ﬁnite subset F of α, and then selecting an element bi ∈ R for each i ∈ F . The ﬁrst choice can be made in α ways, and the second in |R||F | = |R| ways. Thus |M | = α|R| = max(α, |R|). 8. We may take B to the set of “vectors” (ei ) with 1 in position i and zeros elsewhere. Thus there is a basis element for each copy of R, so |B| = α. Section 4.4 1. To prove that the condition is necessary, take the determinant of the equation P P −1 = I. Suﬃciency follows from Cramer’s rule. 2. A homomorphism f : V → W is determined by its action on elements of the form (0, . . . , 0, xj , 0, . . . , 0). Thus we must examine homomorphisms from Vj to ⊕m i=1 Wi . Because of the direct sum, such mappings are assembled from homomorphisms from Vj to Wi , i = 1, . . . , m. Thus f may be identiﬁed with an m × n matrix whose ij element is a homomorphism from Vj to Wi . Formally, we have an abelian group isomorphism HomR (V, W ) ∼ = [HomR (Vj , Wi )]. 3. In Problem 2, replace V and W by V n and take all Wi and Vj to be V . This gives an abelian group isomorphism of the desired form. Now if f corresponds to [fij ] where fij : Vj → Vi , and g corresponds to [gij ], then the composition g ◦ f is assembled from homomorphisms gik ◦fkj : Vj → Vk → Vi . Thus composition of homomorphisms corresponds to multiplication of matrices, and we have a ring isomorphism. 4. In (4.4.1), take n = m = 1 and M = R. 5. Since f (x) = f (x1) = xf (1), we may take r = f (1). 6. This follows from Problems 3 and 4, with V = R. 7. If the endomorphism f is represented by the matrix A and g by B, then for any c ∈ R, we have c(g ◦ f ) = (cg) ◦ f = g ◦ (cf ), so EndR (M ) is an R-algebra. Furthermore, cf is represented by cA, so the ring isomorphism is also an R-module homomorphism, hence an R-algebra isomorphism. Section 4.5 1. Add column 2 to column 1, then add -3 times row 2 to row 3. The Smith normal form is  1 0 S = 0 3 0 0

column 1 to column 2, then add -4 times  0 0 6

page 15 of Solutions Chapters 1-5

2. The matrix P −1 is the product of the elementary column matrices in the order in which they appeared. Thus      1 0 0 1 −3 0 1 −3 0 P −1 =  1 1 0   0 1 0  =  1 −2 0  0 0 1 0 0 1 0 0 1 

−2 P =  −1 0

3 1 0

 0 0 1

The matrix Q is the product of the elementary row matrices in opposite order (i.e., if R1 appears ﬁrst, followed by R2 and R3 , then Q = R3 R2 R1 ). In this case there is only one matrix, so   1 0 0 Q = 0 1 0 0 −4 1 A direct computation shows that QAP −1 = S. 3. The new basis is given by Y = P X, i.e., y1 = −2x1 + 3x2 , y2 = −x1 + x2 , y3 = x3 . The new set of generators is given by V = SY , i.e., v1 = y1 , v2 = 3y2 , v3 = 6y3 . 4. Let di = a1 · · · ai . Then di is the gcd of the i × i minors of S, and hence of A. The ai are recoverable from the di via a1 = d1 and ai = di /di−1 , i > 1. Thus the ai are determined by the matrix A and do not depend on any particular sequence leading to a Smith normal form. 5. If A and B have the same Smith normal form S, then A and B are each equivalent to S and therefore equivalent to each other. If A and B are equivalent, then by the result stated before Problem 4, they have the same gcd of i × i minors for all i. By Problem 4, they have the same invariant factors and hence the same Smith normal form. 6. Here are the results, in sequence: 1. The second row is now (3 2 -13 2) 2. The ﬁrst row is (3 2 -13 2) and the second row is (6 4 13 5) 3. The second row is (0 0 39 1) and the third row is (0 0 51 4) 4. The third row becomes (0 0 12 3) 5. The second row is (0 0 12 3) and the third row is (0 0 39 1) 6. The third row is (0 0 3 -8) 7. The second row is (0 0 3 -8) and the third row is (0 0 12 3) 8. The third row is (0 0 0 35) 9. The ﬁrst row is now (3 2 2 -38) 10. The ﬁnal matrix is   3 2 2 32  0 0 3 27  . 0 0 0 35 7. We see from the Hermite normal form that we can take x = 0, y = 7, z = 9, provided 0 and 35 are congruent mod m. Thus m must be 5,7 or 35. Section 4.6 1. 441 = 32 × 72 , and since there are two partitions of 2, there are 2 × 2 = 4 mutually nonisomorphic abelian groups of order 441, with the following invariant factors: (1) a1 = 32 72 , G ∼ = Z441 (2) a1 = 30 71 , a2 = 32 71 , G ∼ = Z7 ⊕ Z63 (3) a1 = 31 70 , a2 = 31 72 , G ∼ = Z3 ⊕ Z147 (4) a1 = 31 71 , a2 = 31 71 , G ∼ = Z21 ⊕ Z21

page 16 of Solutions Chapters 1-5

2. 40 = 23 × 51 , and since there are three partitions of 3 and one partition of 1, there are 3 × 1 = 3 mutually nonisomorphic abelian groups of order 40, with the following invariant factors: (1) a1 = 23 51 , G ∼ = Z40 (2) a1 = 21 50 , a2 = 22 51 , G ∼ = Z2 ⊕ Z20 (3) a1 = 21 50 , a2 = 21 50 , a3 = 21 51 , G ∼ = Z2 ⊕ Z2 ⊕ Z10 3. The steps in the computation of the Smith normal form are         1 5 3 1 5 3 1 0 0 1 0 0  2 −1 7  →  0 −11 1  →  0 −11 1  →  0 1 −11  3 4 2 0 −11 −7 0 −11 −7 0 −7 −11 

1 → 0 0

0 1 0

  0 1 −11  →  0 −88 0

0 1 0

 0 0  88

Thus G ∼ = Z1 ⊕ Z1 ⊕ Z88 ∼ = Z88 . 4. Cancelling a factor of 2 is not appropriate. After the relations are imposed, the group is no longer free, so that 2y = 0 does not imply that y = 0. Another diﬃculty is that the submodule generated by 2x1 + 2x2 + 8x3 is not the same as the submodule generated by x1 + x2 + 4x3 . 5. Take M = ⊕∞ n=1 Mn , where each Mn is a copy of Z. Take N = Z and P = 0. Since the union of a countably inﬁnite set and a ﬁnite set is still countably inﬁnite, we have the desired result. 6. If N and P are not isomorphic, then the decompositions of N and P will involve diﬀerent sequences of invariant factors. But then the same will be true for M ⊕ N and M ⊕ P , so M ⊕ N and M ⊕ P cannot be isomorphic. Section 4.7 1. If u is another solution, then f u = f u (= vf ), and since f is injective, u = u . 2. By commutativity, wgf a = g vf a, and by exactness, gf = 0. Thus vf a ∈ ker g = im f by exactness. 3. For commutativity, we must have f ua = vf a = f a , so ua = a . Note that a is unique because f is injective. Checking that u is a homomorphism is routine, e.g., if vf ai = f ai , i = 1, 2, then vf (a1 + a2 ) = f (a1 + a2 ), so u(a1 + a2 ) = ua1 + ua2 , etc. 4. uc = ugb = g vb. 5. Suppose c = gb1 = gb2 . Then b1 − b2 ∈ ker g = im f by exactness, so b1 − b2 = f a. Then f wa = vf a = v(b1 − b2 ). By exactness, 0 = g f wa = g v(b1 − b2 ), and the result follows. 6. Add a vertical zero map at the left side of the diagram and apply (ii) of the four lemma. 7. Add a vertical zero map at the right side of the diagram and apply (i) of the four lemma. 8. Add a vertical zero map w at the right side of the diagram and apply (ii) of the four lemma, shifting the notation [s → t, t → u, u → v, v → w]. 9. Since u and g are surjective, v must be also, by commutativity. 10. Since f and u are injective, f t, hence t, must be also, by commutativity. 11. Add a vertical zero map s at the left side of the diagram, and apply (i) of the four lemma, shifting notation [t → s, u → t, v → u, w → v]. 12. If vb = 0, then b = gm, hence 0 = vgm = g um. Thus um ∈ ker g = im f , say um = f a . Since t is surjective, a = ta, so uf a = f ta = f a . Therefore um and uf a are both equal to f a . Since u is injective, m = f a, so b = gm = gf a = 0, proving that v is injective. 13. Let a ∈ A . Since u is surjective, f a = um, so vgm = g um = g f a = 0. Since v is injective, gm = 0, hence m ∈ ker g = im f , so m = f a. Thus um = uf a = f ta. Therefore f a and f ta are both equal to um. Since f is injective, a = ta, proving that t is surjective. Section 5.1

page 17 of Solutions Chapters 1-5

1. The kernel of any homomorphism is a (normal) subgroup. If g ∈ ker Φ then g(xH) = xH for every x ∈ G, so by (1.3.1), x−1 gx ∈ H. Take x = g to get g ∈ H. 2. By Problem 1, ker Φ is a normal subgroup of G, necessarily proper since it is contained in H. Since G is simple, ker Φ = {1}, and hence Φ is injective. Since there are n left cosets of H, Φ maps into Sn . 3. If [G : H] = n < ∞, then by Problem 2, G can be embedded in Sn , so G is ﬁnite, a contradiction. 4. g(xH) = xH iﬀ x−1 gx ∈ H iﬀ g ∈ xHx−1 . 5. If x ∈ G, then K = xKx−1 ⊆ xHx−1 , and since x is arbitrary, K ⊆ N . 6. g1 (H ∩ K) = g2 (H ∩ K) iﬀ g2−1 g1 ∈ H ∩ K iﬀ g1 H = g2 H and g1 K = g2 K, proving both assertions. 7. Since [G : H] and [G : K] are relatively prime and divide [G : H ∩ K] by (1.3.5), their product divides, and hence cannot exceed, [G : H ∩ K]. 8. By the ﬁrst isomorphism theorem, G/N is isomorphic to a group of permutations of L, the set of left cosets of H. But |L| = [G : H] = n, so by Lagrange’s theorem, |G/N | divides |Sn | = n!. 9. Since n > 1, H is a proper subgroup of G, and since N is a subgroup of H, N is a proper subgroup of G as well. If N = {1}, then |G| = [G : N ], so by Problem 8, G divides n!, contradicting the hypothesis. Thus {1} < N < G, and G is not simple. Section 5.2 1. For arbitrary σ and π, we have πσπ −1 (π(i)) = πσ(i). In the cycle decomposition of σ, i is followed by σ(i), and in the cycle decomposition of πσπ −1 , π(i) is followed by πσ(i), exactly as in the given numerical example. 2. If g ∈ CG (S) and x ∈ S then gxg −1 = x, so gSg −1 = S, hence CG (S) ≤ NG (S). If g ∈ NG (S) and x ∈ S, then gxg −1 ∈ S, and the action is legal. As in (5.1.3), Example 3, the kernel of the action consists of all elements of NG (S) that commute with everything in S, that is, NG (S) ∩ CG (S) = CG (S). 3. We have z ∈ G(gx) iﬀ zgx = gx iﬀ g −1 zgx = x iﬀ g −1 zg ∈ G(x) iﬀ z ∈ gG(x)g −1 . 4. We have g1 G(x) = g2 G(x) iﬀ g2−1 g1 ∈ G(x) iﬀ g2−1 g1 x = x iﬀ g1 x = g2 x, proving that Ψ is well-deﬁned and injective. If the action is transitive and y ∈ X, then for some x, y = gx = Ψ(gG(x)) and Ψ is surjective. 5. If g, h ∈ G, then h takes gx to hgx. In the coset action, the corresponding statement is that h takes gG(x) to hgG(x). The formal statement is that Ψ is a“G-set isomorphism”. In other words, Ψ is a bijection of the space of left cosets of G(x) and X, with Ψ(hy) = hΨ(y) for all h ∈ G and y in the coset space. Equivalently, the following diagram is commutative. gx Ψ↑

→

hgx Ψ↑

gG(x) → hgG(x) 6. The two conjugacy classes are {1} and G \ {1}. Thus if |G| = n > 1, the orbit sizes under conjugacy on elements are 1 and n − 1. But each orbit size divides the order of the group, so n − 1 divides n. Therefore n = k(n − 1), where k is a positive integer. Since k = 1 is not possible, we must have k ≥ 2, so n ≥ 2(n − 1), so n ≤ 2. 7. If gi is an element in the ith conjugacy class, 1 ≤ i ≤ k, then by the orbit-stabilizer theorem, the size of this class is |G|/|CG (gi )|. Since the orbits partition G, the sum of the class sizes is |G|, and k 1 =1 x i=1 i where xi = |CG (gi )|. If, say, g1 = 1, so that x1 = |G|, the result follows from the observation that each xi , in particular x1 , is bounded by N (k).

page 18 of Solutions Chapters 1-5

Section 5.3 1. The group elements are I, R = (1, 2, 3, 4), R2 = (1, 3)(2, 4), R3 = (1, 4, 3, 2), F = (1)(3)(2, 4), RF = (1, 2)(3, 4), R2 F = (1, 3)(2)(4), R3 F = (1, 4)(2, 3). Thus the number of distinct colorings is 1 4 1 (n + n + n2 + n + n3 + n2 + n3 + n2 ) = (n4 + 2n3 + 3n2 + 2n). 8 8 2. Yes. If the vertices of the square are 1,2,3,4 in counterclockwise order, we can identify vertex 1 with side 12, vertex 2 with side 23, vertex 3 with side 34, and vertex 4 with side 41. This gives a one-to-one correspondence between colorings in one problem and colorings in the other. 3. If the vertices of the square are 1,2,3,4 in counterclockwise order, then W GGW will mean that vertices 1 and 4 are colored white, and vertices 2 and 3 green. The equivalence classes are {W W W W }, {GGGG}, {W GGG, GW GG, GGW G, GGGW }, {GW W W, W GW W, W W GW, W W W G}, {W W GG, GW W G, GGW W, W GGW }, {W GW G, GW GW }. 4. Label (-1,0) as vertex 1, (0,0) as vertex 2, and (1,0) as vertex 3. Then I = (1)(2)(3) and σ = (1, 3)(2). Thus the number of distinct colorings is 12 (n3 + n2 ). 5. We have free choice of color in two cycles of I and one cycle of σ. The number of distinct colorings is 12 (n2 + n). 6. We can generate a rotation by choosing a face of the tetrahedron to be placed on a table or other ﬂat surface, and then choosing a rotation of 0,120 or 240 degrees. Thus there are 12 rotations, and we have enumerated all of them. By examining what each rotation does to the vertices, we can verify that all permutations are even. Since A4 has 4!/2 = 12 members, G must coincide with A4 , up to isomorphism. 7. The members of A4 are (1,2,3), (1,3,2), (1,2,4), (1,4,2), (1,3,4), (1,4,3), (2,3,4), (2,4,3), (1,2)(3,4), (1,3)(2,4), (1,4)(2,3), and the identity. Counting cycles of length 1, we have 11 permutations with 2 cycles and one permutation with 4 cycles. The number of distinct 1 colorings is 12 (n4 + 11n2 ). 8. In the above list of permutations, the ﬁrst 8 have no ﬁxed colorings. In the next 3, we can pick a cycle to be colored B, and pick a diﬀerent color for the other cycle. This gives 2×3 = 6 ﬁxed colorings. For the identity, we can pick two vertices to be colored B, and then choose

4 2 a diﬀerent color for each of the other two vertices. The number of ﬁxed colorings is 2 3 = 54. The number of distinct colorings of the vertices is (6+54)/12=5. 9. As in Problem 6, a rotation can be generated by choosing a face of the cube to be placed on a table, and then choosing a rotation of 0,±90 or 180 degrees. Thus there are 24 rotations, and we have enumerated all of them. Alternatively, there is a one-to-one correspondence between rotations and permutations of the 4 diagonals of the cube. Since there are 4! = 24 permutations of a set with 4 elements, there can be no additional rotations. The correspondence between rotations and permutations of the diagonals yields an isomorphism of G and S4 . 10. Any permutation of the faces except the identity has a cycle of length 2 or more, and each of the faces within that cycle must receive the same color, which is a contradiction. Thus f (π) = 0 for π = I. Now I ﬁxes all legal colorings, and since there are 6 colors and 6 faces, the number of legal colorings is 6! = 720. The number of distinct colorings is therefore 720/24=30. Remark This problem can be solved directly without using the heavy machinery of this section. Without loss of generality, choose any particular color for a particular face, and

page 19 of Solutions Chapters 1-5

move the cube so that this face is at the bottom. Choose one of the remaining 5 colors for the top face. The number of allowable colorings of the 4 remaining sides of the cube is the number of circular permutations of 4 objects, which is 3! = 6. The number of distinct colorings is 5 × 6 = 30. 11. The group G = {1, R, R2 , . . . , Rp−1 } is cyclic of order p. Since p is prime, each Ri , i = 1, . . . , p − 1, has order p, and therefore as a permutation of the vertices consists of a single cycle. Thus the number of distinct colorings is 1 p [n + (p − 1)n]. p 12. Since the result of Problem 11 is an integer, np + (p − 1)n = np − n + np is a multiple of p, hence so is np − n. Thus for any positive integer n, np ≡ n mod p. It follows that if n is not a multiple of p, then np−1 ≡ 1 mod p. Section 5.4 1. Let G act on subgroups by conjugation. If P is a Sylow p-subgroup, then the stabilizer of P is NG (P ) (see (5.2.2), Example 4). By (5.2.3), the index of NG (P ) is np . 2. Since P is normal in NG (P ) (see (5.2.2), Example 4), P Q = QP ≤ G by (1.4.3). By (5.2.4), P Q is a p-subgroup. 3. The Sylow p-subgroup P is contained in P Q, which is a p-subgroup by Problem 2. Since a Sylow p-subgroup is a p-subgroup of maximum possible size, we have P = P Q, and therefore Q ⊆ P. 4. (a) By deﬁnition of normalizer, we have gP g −1 ≤ gNG (P )g −1 ≤ gHg −1 = H. Thus P and gP g −1 are subgroups of H, and since they are p-subgroups of maximum possible size, they are Sylow p-subgroups of H. (b) Since H is always a subgroup of its normalizer, let g ∈ NG (H). By (a), P and gP g −1 are conjugate in H, so for some h ∈ H we have gP g −1 = hP h−1 . Thus (h−1 g)P (h−1 g)−1 = P , so h−1 g ∈ NG (P ) ≤ H. But then g ∈ H, and the result follows. 5. By the second isomorphism theorem, [N : P ∩ N ] = [P N : P ] = |P N |/|P |. Since |P | is the largest possible power of p for p-subgroups of G, [P N : P ] and p must be relatively prime. Therefore [N : P ∩ N ] and p are relatively prime, so P ∩ N is a p-subgroup of N of maximum possible size, i.e., a Sylow p-subgroup of N . 6. By the third isomorphism theorem, [G/N : P N/N ] = [G : P N ] = |G|/|P N |. Since |G|/|P | and p are relatively prime and P ≤ P N , it follows that |G|/|P N | and p are relatively prime. The result follows as in Problem 5. 7. Since f is an automorphism, f (P ) is a subgroup of G and has the same number of elements as P , in other words, f (P ) is a Sylow p-subgroup. By hypothesis, f (P ) = P . 8. By (1.3.5), [G : N ] = [G : H][H : N ] = p[H : N ], and since [G : N ] divides p! = p(p − 1)!, the result follows. 9. If q is a prime factor of [H : N ], then by Problem 8, q is a divisor of some integer between 2 and p − 1, in particular, q ≤ p − 1. But by Lagrange’s theorem, q divides |H|, hence q divides |G|. This contradicts the fact that p is the smallest prime divisor of |G|. We conclude that there are no prime factors of [H : N ], which means that [H : N ] = 1, Thus H = N , proving that H is normal in G. Section 5.5 1. This follows from (5.5.6), part (iii), with p = 3 and q = 5. 2. Let Z(G)a be a generator of G/Z(G). If g1 , g2 ∈ G, then Z(G)g1 = Z(G)ai for some i, so g1 a−i = z1 ∈ Z(G), and similarly g2 a−j = z2 ∈ Z(G). Thus g1 g2 = ai z1 aj z2 = z1 z2 ai+j = z2 z1 aj+i = z2 aj z1 ai = g2 g1 . 3. By (5.5.3), the center Z(G) is nontrivial, so has order p or p2 . In the latter case, G = Z(G), so G is abelian. If |Z(G)| = p, then |G/Z(G)| = p, and G/Z(G) has order p and is therefore cyclic. By Problem 2, G is abelian (and |Z(G)| must be p2 , not p).

page 20 of Solutions Chapters 1-5

4. Each Sylow p-subgroup is of order p and therefore has p − 1 elements of order p, with a similar statement for q and r. If we include the identity, we have 1 + np (p − 1) + nq (q − 1) + nr (r − 1) distinct elements of G, and the result follows. 5. G cannot be abelian, for if so it would be cyclic of prime order. By (5.5.5), np , nq and nr are greater than 1. We know that np divides qr and np > 1. But np can’t be q since q ≡ 1 mod p (because p > q). Similarly, np can’t be r, so np = qr. Now nq divides pr and is greater than 1, so as above, nq must be either p or pr (it can’t be r because q > r, so r ≡ 1 mod q). Thus nq ≥ p. Finally, nr divides pq and is greater than 1, so nr is p, q, or pq. Since p > q, we have nr ≥ q. 6. Assume that G is simple. Substituting the inequalities of Problem 5 into the identity of Problem 4, we have pqr ≥ 1 + qr(p − 1) + p(q − 1) + q(r − 1). Thus 0 ≥ pq − p − q + 1 = (p − 1)(q − 1), a contradiction. 7. Since |P | = pr with r ≥ 1 and m > 1, we have 1 < |P | < |G|. Since G is simple, P is not normal in G. By (5.5.4), n > 1. By Problem 9 of Section 5.1, |G| divides n!. 8. Assume G simple, and let n = np with p = 5. By Sylow (2), n divides 24 = 16 and n ≡ 1 mod 5. The only divisors of 16 that are congruent to 1 mod 5 are 1 and 16, and 1 is excluded by Problem 7. Thus the only possibility is n = 16, and by Problem 7, 24 56 divides 16!, hence 56 divides 16!. But in the prime factorization of 16!, 5 appears with exponent 3 (not 6), due to the contribution of 5,10 and 15. We have reached a contradiction, so G cannot be simple. Section 5.6 1. Apply the Jordan-H¨ older theorem to the series 1 N G. 2. Z has no composition series. By Section 1.1, Problem 6, each nontrivial subgroup of Z consists of multiples of some positive integer, so the subgroup is isomorphic to Z itself. Thus Z has no simple subgroups, so if we begin with {0} and attempt to build a composition series, we cannot even get started. 3. We have the composition series 1 ? Z2 ? Z2 ⊕ Z3 ∼ = Z6 (or 1 ? Z3 ? Z2 ⊕ Z3 ∼ = Z6 ) and 1 ? A3 ? S3 . 4. An consists of products of an even number of transpositions, and the result follows from the observation that (a, c)(a, b) = (a, b, c) and (c, d)(a, b) = (a, d, c)(a, b, c). 5. If (a, b, c) ∈ N and (d, e, f ) is any 3-cycle, then for some permutation π we have π(a, b, c)π −1 = (d, e, f ). Explicitly, we can take π(a) = d, π(b) = e, π(c) = f ; see Section 5.2, Problem 1. We can assume without loss of generality that π is even, for if it is odd, we can replace it by (g, h)π, where g and h are not in {d, e, f }. (We use n ≥ 5 here.) Since N is normal, (d, e, f ) ∈ N . 6. If N contains (1,2,3,4), then it contains (1,2,3)(1,2,3,4)(1,3,2)=(1,4,2,3), and hence contains (1,4,2,3)(1,4,3,2)=(1,2,4), contradicting Problem 5. If N contains (1,2,3,4,5), then it contains (1,2,3)(1,2,3,4,5)(1,3,2)=(1,4,5,2,3), and so contains (1,4,5,2,3)(1,5,4,3,2)=(1,2,4), a contradiction. The analysis for longer cycles is similar. [Actually, we should have assumed that N contains a permutation π whose disjoint cycle decomposition is · · · (1, 2, 3, 4) · · ·. But multiplication by π −1 = · · · (1, 4, 3, 2) · · · cancels the other cycles.] 7. If N contains (1,2,3)(4,5,6), then it must also contain (3,4,5)(1,2,3)(4,5,6)(3,5,4)=(1,2,4)(3,6,5). Thus N also contains (1,2,4)(3,6,5)(1,2,3)(4,5,6)=(1,4,3,2,6), which contradicts Problem 6. [If the decomposition of a permutation σ in N contains a single 3-cycle, then σ 2 is a 3-cycle in N , because a transposition is its own inverse. This contradicts Problem 5.] 8. If, (1, 2)(3, 4) ∈ N , then (1,5,2)(1,2)(3,4)(1,2,5)=(1,5)(3,4) belongs to N , and so does (1,5)(3,4)(1,2)(3,4)=(1,2,5), contradicting Problem 5. 9. If N contains (1,2)(3,4)(5,6)(7,8), then it contains

page 21 of Solutions Chapters 1-5

(2,3)(4,5)(1,2)(3,4)(5,6)(7,8)(2,3)(4,5)=(1,3)(2,5)(4,6)(7,8). Therefore N contains (1,3)(2,5)(4,6)(7,8)(1,2)(3,4)(5,6)(7,8)=(1,5,4)(2,3,6), contradicting Problem 7. 10. We can reproduce the analysis leading to the Jordan-H¨ older theorem, with appropriate notational changes. For example, we replace the “subnormal” condition Gi Gi+1 by the “normal” condition Gi G. 11. We say that N is a minimal normal subgroup of H if {1} < N H and there is no normal subgroup of H strictly between {1} and N . In a chief series, there can be no normal subgroup of G strictly between Gi and Gi+1 . Equivalently, by the correspondence theorem, there is no normal subgroup of G/Gi strictly between Gi /Gi = {1} and Gi+1 /Gi . Thus Gi+1 /Gi is a minimal normal subgroup of G/Gi . Section 5.7 1. S3 is nonabelian and solvable (1 ? A3 ? S3 ). 2. Let 1 = G0 ? G1 ? · · · ? Gr = G be a composition series, with all Gi /Gi−1 cyclic of prime order (see (5.7.5)). Since |Gi | = |Gi /Gi−1 ||Gi−1 | and G0 is ﬁnite, an induction argument shows that G is ﬁnite. 3. The factors of a composition series are simple p-groups P , which must be cyclic of prime order. For 1 ? Z(P ) P , so Z(P ) = P and P is abelian, hence cyclic of prime order by (5.5.1). 4. S3 is solvable by Problem 1, but is not nilpotent. Since S3 is nonabelian, a central series must be of the form 1 ? H ? S3 with H ⊆ Z(S3 ) = 1, a contradiction. 5. If Sn is solvable, then so is An by (5.7.4), and this contradicts (5.7.2). 6. By (5.5.3), P has a nontrivial center. Since Z(P ) is normal in P and P is simple, Z(P ) = P and P is abelian. By (5.5.1), P is cyclic of prime order, and since P is a p-group, the only possibility is |P | = p. 7. Let N be a maximal proper normal subgroup of P . (N exists because P is ﬁnite and nontrivial, and 1 ? P .). Then the p-group P/N is simple (by the correspondence theorem). By Problem 6, |P/N | = p. 8. If P is a Sylow p-subgroup of G, then |P | = pr and by Problem 7, P has a subgroup Q1 of index p, hence of order pr−1 . If Q1 is nontrivial, the same argument shows that Q1 has a subgroup Q2 of order pr−2 . An induction argument completes the proof. 9. Let G = D6 , the group of symmetries of the equilateral triangle. Take N = {I, R, R2 }, where R is rotation by 120 degrees. Then N has index 2 in G and is therefore normal. Also, N has order 3 and G/N has order 2, so both N and G/N are cyclic, hence abelian. But G is not abelian, since rotations and reﬂections do not commute. 10. It follows from the splicing technique given in the proof of (5.7.4) that dl(G) ≤ dl(N) + dl(G/N). Section 5.8 1. Let H =< a | an >, and let Cn be a cyclic group of order n generated by a. Then an = 1 in Cn , and since an+j = aj , we have |H| ≤ n = |Cn |. The result follows as in (5.8.6). 2. The discussion in Example 4 of (2.1.3), with i = a and j = b, shows that the quaternion group Q satisﬁes all the relations. Since ab = ba−1 , it follows as in (1.2.4) that every element of the given presentation H is of the form br as , r, s ∈ Z. Since b2 = a2 , we can restrict r to 0 or 1, and since a4 = 1, we can restrict s to 0,1,2 or 3. Thus |H| ≤ 8, and the result follows as in (5.8.6). 3. Take a = (1, 2, 3) and b = (1, 2) to show that S3 satisﬁes all the relations. Since ba = a−1 b, each element of H is of the form ar bs , r, s ∈ Z. Since a3 = b2 = 1, |H| ≤ 3 × 2 = 6, and the result follows as in (5.8.6). 4. No. There are many counterexamples; an easy one is Cn =< a | an = 1, a2n = 1 >, the cyclic group of order n. −1 5. n−1 2 n1 = h2 h1 ∈ N ∩ H = 1. 6. Take ψ to be the inclusion map. Then πψ(h) = π(h) = π(1h) = h. To show that π is a −1 homomorphism, note that n1 h1 n2 h2 = n1 (h1 n2 h−1 1 )h1 h2 and h1 n2 h1 ∈ N .

page 22 of Solutions Chapters 1-5

7. If g ∈ G then g = (gπ(g)−1 )π(g) with π(g) ∈ H and π(gπ(g)−1 ) = π(g)π(g)−1 = 1, so gπ(g)−1 ∈ N . [Remember that since we are taking ψ to be inclusion, π is the identity on H.] Thus G = N H. If g ∈ N ∩ H, then g ∈ ker π and g ∈ H, so g = π(g) = 1, proving that H ∩ N = 1. 8. If we deﬁne π(n, h) = (1, h), i(n, 1) = (n, 1), and ψ(1, h) = (1, h), then the sequence of Problem 6 is exact and splits on the right. 9. We have (n1 h1 )(n2 h2 ) = n1 (h1 n2 h−1 1 )h1 h2 , so we may take f (h) to be the inner automorphism of N given by conjugation by h ∈ H. 10. Consider the sequence 1

→

C3

i →

S3

π →

C2

→

1

where C3 consists of the identity 1 and the 3-cycles (1,2,3) and (1,3,2), and C2 consists of the identity and the 2-cycle (1,2). The map i is inclusion, and π takes each 2-cycle to (1,2) and each 3-cycle to the identity. The identity map from C2 to S3 gives a right-splitting, but there is no left splitting. If g were a left-splitting map from S3 to C3 , then g(1, 2) = (1, 2, 3) is not possible because g(1) = g(1, 2)g(1, 2) = (1, 2, 3)(1, 2, 3) = (1, 3, 2), a contradiction. Similarly, g(1, 2) = (1, 3, 2) is impossible, so g(1, 2) = 1, so g ◦ i cannot be the identity. Explicitly, g(2, 3) = g((1, 2)(1, 2, 3)) = g(1, 2, 3) = (1, 2, 3), and g(1, 3) = g((1, 2)(1, 3, 2)) = g(1, 3, 2) = (1, 3, 2). Consequently, g(1, 3, 2) = g((1, 3)(2, 3)) = 1, a contradiction. 11. In the exact sequence of Problem 6, take G = Zp2 , N = Zp , H = G/N ∼ = Zp , i the inclusion map, and π the canonical epimorphism. If f is a right-splitting map, its image must be a subgroup with p elements (since f is injective), and there is only one such subgroup, namely Zp . But then π ◦ f = 0, a contradiction. 12. If g ∈ G, then gP g −1 ⊆ gN g −1 = N , so P and gP g −1 are both Sylow p-subgroups of N . By Sylow (3), they are conjugate in N (the key point). Thus for some n ∈ N we have P = n(gP g −1 )n−1 . But then by deﬁnition of normalizer we have ng ∈ NG (P ), hence g ∈ N NG (P ). 13. The multiplication table of the group is completely determined by the relations an = 1, b2 = 1, and bab−1 = a−1 . The relations coincide with those of D2n , with a = R and b = F. 14. The relation an = 1 disappears, and we have < a, b | b2 = 1, bab−1 = a−1 >.

page 1 of Solutions Chapters 6-10

Section 6.1 1. We have r1 = 2, r2 = 1, r3 = 1 so t1 = 1, t2 = 0, t3 = 1. The algorithm terminates in one step after after subtraction of (X1 + X2 + X3 )(X1 X2 X3 ). The given polynomial can be expressed as e1 e3 . 2. We have r1 = 2, r2 = 1, r3 = 0 so t1 = 1, t2 = 1, t3 = 0. At step 1, subtract (X1 + X2 + X3 )(X1 X2 + X1 X3 + X2 X3 ). The result is −3X1 X2 X3 + 4X1 X2 X3 = X1 X2 X3 . By inspection (or by a second step of the algorithm), the given polynomial can be expressed as e1 e2 + e3 . 3. Equation (1) follows upon taking σ1 (h) outside the summation and using the linear dependence. Equation (2) is also a consequence of the linear dependence, because σi (h)σi (g) = σi (hg). 4. By hypothesis, the characters are distinct, so for some h ∈ G we have σ1 (h) = σ2 (h). Thus in (3), each ai is nonzero and σ1 (h) − σi (h)

=0 = 0

if i = 1; if i = 2.

This contradicts the minimality of r. (Note that the i = 2 case is important, since there is no contradiction if σ1 (h) − σi (h) = 0 for all i.) 5. By (3.5.10), the Galois group consists of the identity alone. Since the identity ﬁxes all √ elements, the ﬁxed ﬁeld of G is Q( 3 2). 6. Since C = R[i], an R-automorphism σ of C is determined by its action on i. Since σ must permute the roots of X 2 + 1 by (3.5.1), we have σ(i) = i or −i. Thus the Galois group has two elements, the identity automorphism and complex conjugation. 7. The complex number z is ﬁxed by complex conjugation if and only if z is real, so the ﬁxed ﬁeld is R. Section 6.2 p 1. The right side is a subset of the left since both Ei and Ei+1 are contained in Ei+1 . p Since Ei is contained in the set on the right, it is enough to show that αi+1 ∈ Ei (Ei+1 ). p By hypothesis, αi+1 is separable over F , hence over Ei (αi+1 ). By Section 3.4, Problem 3, p p ) ⊆ Ei (Ei+1 ). αi+1 ∈ Ei (αi+1 p ), to conclude 2. Apply Section 3.4, Problem 7, with E = F (E p ) replaced by Ei+1 = Ei (Ei+1 that Ei+1 is separable over Ei . By the induction hypothesis, Ei is separable over F . By transitivity of separable extensions (Section 3.4, Problem 8), Ei+1 is separable over F . By induction, E/F is separable. 3. Let fi be the minimal polynomial of αi over F . Then E is a splitting ﬁeld for f = f1 · · · fn over F , and the result follows. 4. This is a corollary of part 2 of the fundamental theorem, with F replaced by Ki−1 and G replaced by Gal(E/Ki−1 ) = Hi−1 . 5. E(A) is a ﬁeld containing E ≥ F and A, hence E(A) contains E and K, so that by deﬁnition of composite, EK ≤ E(A). But any ﬁeld (in particular EK) that contains E and K contains E and A, hence contains E(A). Thus E(A) ≤ EK. 6. If σ ∈ G, deﬁne Ψ(σ)(τ (x)) = τ σ(x), x ∈ E. Then ψ(σ) ∈ G . [If y = τ (x) ∈ F with x ∈ F , then Ψ(σ)y = Ψ(σ)τ x = τ σ(x) = τ (x) = y.] Now Ψ(σ1 σ2 τ (x)) = τ σ1 σ2 (x) and Ψ(σ1 )Ψ(σ2 )τ (x) = Ψ(σ1 )τ σ2 (x) = τ σ1 σ2 (x), so Ψ is a group homomorphism. The inverse of Ψ is given by Ψ (σ )τ −1 y = τ −1 σ (y), σ ∈ G , y ∈ E . To see this, we compute

Ψ (Ψ(σ))τ −1 y = τ −1 Ψ(σ)y = τ −1 Ψ(σ)τ x = τ −1 τ σ(x) = σ(x) = σ(τ −1 y). Thus Ψ Ψ is the identity on G. 7. Since H is a normal subgroup of G, its ﬁxed ﬁeld L = F(H ) is normal over F , so by minimality of the normal closure, we have N ⊆ L. But all ﬁxed ﬁelds are subﬁelds of N , so

page 2 of Solutions Chapters 6-10

L ⊆ N , and consequently L = N . 8. If σ ∈ H , then σ ﬁxes everything in the ﬁxed ﬁeld N , so σ is the identity. Thus the largest normal subgroup of G that is contained in H is trivial. But this largest normal subgroup is the core of H in G, and the resulting formula follows from Problems 4 and 5 of Section 5.1. Section 6.3 1. G = {σ1 , . . . , σn } where σi is the unique F -automorphism of E that takes α to αi . 2. We must ﬁnd an α such that 1, α, . . . , αn−1 is a basis for E/Q. If α = b1 x1 +· · ·+bn xn , we can compute the various powers of α and write αi = ci1 x1 + · · · + cin xn , i = 0, 1, . . . , n − 1, where each cij is a rational number. The powers of α will form a basis iﬀ det[cij ] = 0. This will happen “with probability 1”; if a particular choice of the bi yields det[cij ] = 0, a slight perturbation of the bi will produce a nonzero determinant. 3. By (6.3.1), we may regard G as a group of permutations of the roots α1 , . . . , αn of f , and therefore G is isomorphic to a subgroup H of Sn . Since G acts transitively on the αi (see (6.3.1)), the natural action of H on {1, 2, . . . , n} is transitive. [For an earlier appearance of the natural action, see the discussion before (5.3.1).] 4. The Galois group G must be isomorphic to a transitive subgroup of S2 , which is cyclic of order 2. There is only one transitive subgroup of S2 , namely S2 itself, so G is a cyclic group of order √ 2. √ √ √ 5. Since [Q( 2, the Galois group G has order 4. √ 2) : Q] √ = 2 and [Q( 2,√ 3) : Q( 2)] = √ [Note that 3 ∈ / Q( 2) because a + b 2 can never be 3 for a, b ∈ Q.] An automorphism √ √ √ √ σ in G must take 2 to ± 2 and 3 to ± 3. Thus σ is either the identity or has order 2. Now a group in which every element has order 1 or 2 must be abelian, regardless of the size of the group [(ab)(ab) = 1, so ab = b−1 a−1 = ba]. Since G is not cyclic, it must be isomorphic to the four group Z2 ⊕ Z2 . (See the analysis in (4.6.4).) 6. Let H be the subgroup generated by H1 and H2 , that is, by H1 ∪H2 . If σ ∈ H1 ∪H2 , then σ ﬁxes K1 ∩ K2 = K. Since H consists of all ﬁnite products (= compositions) of elements in H1 or H2 , everything in H ﬁxes K, so that K ⊆ F(H). On the other hand, if x ∈ F(H) / F(H), a contradiction. but x ∈ / K, say x ∈ / K1 . Then some τ ∈ H1 ⊆ H fails to ﬁx x, so x ∈ Therefore K = F(H). 7. The ﬁxed ﬁeld is K1 K2 , the composite of K1 and K2 . For if σ ﬁxes K1 K2 , then it ﬁxes both K1 and K2 , so σ belongs to H1 ∩ H2 . Conversely, if σ ∈ H1 ∩ H2 , then σ is the identity on both K1 and K2 . But by the explicit form of K1 K2 (see Section 3.1, Problem 1 and Section 6.2, Problem 5), σ is the identity on K1 K2 . Thus F(H1 ∩ H2 ) = K1 K2 . 8. We have E = F (α1 , . . . , αn ), where the αi are the roots of f . Since min(αi , F ) divides the separable polynomial f , each αi is separable over F . By Section 6.2, Problem 1, E is separable over F . 9. Since [Q(θ, i)/Q] = [Q(θ)/Q][Q(θ, i)/Q(θ)] = 4 × 2 = 8, we have |G| = 8. Any σ ∈ G must map θ to a root of f (4 choices), and i to a root of X 2 + 1 (2 choices, i or −i). Since σ is determined by its action on θ and i, we have found all 8 members of G. 10. Let σ(θ) = iθ, σ(i) = i, and let τ (θ) = θ, τ (i) = −i. Then σ 4 = 1, τ 2 = 1, and the automorphisms 1, σ, σ 2 , σ 3 , τ, στ, σ 2 τ, σ 3 τ are distinct (by direct veriﬁcation). Also, we have group in Section 5.8. στ = τ σ −1 = τ σ 3 . The result follows from the analysis of the dihedral √ 11. By direct veriﬁcation, every member of N ﬁxes iθ2 = i 2. Since N has √ index 2 in G, the ﬁxed ﬁeld of N has degree 2 over√Q. But the minimal polynomial of i 2 over Q is X 2 + 2, and it follows that F(N ) = Q(i 2}. F(N ) is the splitting ﬁeld of X 2 + 2 over Q and is therefore normal over Q, as predicted by Galois theory. Section 6.4 1. We have α4 = 1 + α + α2 + α3 and α5 = 1. Thus the powers of α do not exhaust the nonzero elements of GF (16). 2. We may assume that E = GF (pn ) and that E contains F = GF (pm ), where n = md. Then [E : F ] = [E : Fp ]/[F : Fp ] = n/m = d. Since E/F is separable, we have E = F (α) by

page 3 of Solutions Chapters 6-10

the theorem of the primitive element. The minimal polynomial of α over F is an irreducible polynomial of degree d. 3. Exactly as in (6.4.5), carry out a long division of X n − 1 by X m − 1. The division will be successful iﬀ m divides n. 4. Since the bi belong to L, we have K ⊆ L, and since h ∈ L[X], it follows that g|h. But g ∈ K[X] by deﬁnition of K, so h|g. Since g and h are monic, they must be equal. In particular, they have the same degree, so [E : L] = [E : K]. Since K ⊆ L, we have L = K. 5. Since L = K, L is completely determined by g. But if f =min(α, F ), then g divides f . Since f has only ﬁnitely many irreducible divisors, there can only be ﬁnitely many intermediate ﬁelds L. 6. Since there are ﬁnitely many intermediate ﬁelds between E and F , the same is true between L and F . By induction hypothesis, L = F (β) for some β ∈ L. Thus E = L(αn ) = F (β, αn ). 7. By hypothesis, there are only ﬁnitely many ﬁelds of the form F (cβ + αn ), c ∈ F . But there are inﬁnitely many choices of c, and the result follows. 8. Since E = F (β, αn ), it suﬃces to show that β ∈ F (cβ + αn ). This holds because β=

(cβ + αn ) − (dβ + αn ) . c−d

9. Let σ : F → F be the Frobenius automorphism, given by σ(x) = xp . Let f = min(α, Fp ) and g = min(αp , Fp ). Then f (αp ) = f (σ(α)) = σ(f (α)) since σ is a monomorphism, and σ(f (α)) = σ(0) = 0. Thus g divides the monic irreducible polynomial f , so g = f . 10. By Problem 9, the subsets are {0}, {1, 3, 9}, {2, 6, 5}, {4, 12, 10}, and {7, 8, 11}. [For example, starting with 2, we have 2 × 3 = 6, 6 × 3 = 18 ≡ 5 mod 13, 5 × 3 = 15 ≡ 2 mod 13.] In the second case, we get {0}, {1, 2, 4, 8}, {3, 6, 9, 12}, {5, 10}, {7, 14, 13, 11}. Section 6.5 1. Ψn (X p ) = i (X p − ωi ) where the ωi are the primitive nth roots of unity. But the roots of X p − ωi are the pth roots of ωi , which must be primitive npth roots of unity because p is prime and p divides n. The result follows. (The map θ → θp is a bijection between primitive npth roots of unity and primitive nth roots of unity, because ϕ(np) = ϕ(n).) 2. By (6.5.1) and (6.5.6), the Galois group of the nth cyclotomic extension of Q can be identiﬁed with the group of automorphisms of the cyclic group of nth roots of unity. By (6.5.6), the Galois group is isomorphic to Un , and the result follows. 3. The powers of 3 mod 7 are 3, 9 ≡ 2, 6, 18 ≡ 4, 12 ≡ 5, 1. 4. This follows from Problem 3 and (1.1.4). 5. σ6 (ω + ω 6 ) = ω 6 + ω 36 = ω + ω 6 , so ω + ω 6 ∈ K. Now ω + ω 6 = ω + ω −1 = 2 cos 2π/7, so ω satisﬁes a quadratic equation over Q(cos 2π/7). By (3.1.9), [Q7 : Q] = [Q7 : K][K : Q(cos 2π/7)][Q(cos 2π/7) : Q] where the term on the left is 6, the ﬁrst term on the right is | < σ6 > | = 2, and the second term on the right is (by the above remarks) 1 or 2. But [K : Q(cos 2π/7)] cannot be 2 (since 6 is not a multiple of 4), so we must have K = Q(cos 2π/7). 6. σ2 (ω + ω 2 + ω 4 ) = ω 2 + ω 4 + ω 8 = ω + ω 2 + ω 4 , so ω + ω 2 + ω 4 ∈ L; σ3 (ω + ω 2 + ω 4 ) = / Q. [If ω 3 +ω 5 +ω 6 = ω +ω 2 +ω 4 , ω 3 +ω 6 +ω 12 = ω 3 +ω 5 +ω 6 = ω +ω 2 +ω 4 , so ω +ω 2 +ω 4 ∈ then we have two distinct monic polynomials of degree 6 satisﬁed by ω (the other is Ψ7 (X)), which is impossible.] 7. By the fundamental theorem, [L : Q] = [G :< σ2 >] = 2, so we must have L =

page 4 of Solutions Chapters 6-10

Q(ω + ω 2 + ω 4 ). 8. The roots of Ψq are the pr th roots of unity that are not pr−1 th roots of unity. Thus r

tp − 1 Xp − 1 Ψq (X) = pr−1 = t−1 X −1 and the result follows. 9. By Problem 1, Ψ18 (X) = Ψ(3)(6) (X) = Ψ6 (X 3 ) = X 6 − X 3 + 1. Section 6.6 1. f is irreducible by Eisenstein, and the Galois group is S3 . This follows from (6.6.7) or via the discriminant criterion of (6.6.3); we have D(f ) = −27(4) = −108, which is not a square in Q. 2. f is irreducible by the rational root test, and D(f ) = −4(−3)3 − 27 = 108 − 27 = 81, a square in Q. Thus the Galois group is A3 . 3. f is irreducible by Eisenstein. The derivative is f (X) = 5X 4 − 40X 3 = 5X 3 (X − 8). We have f (x) positive for x < 0 and for x > 8, and f (x) negative for 0 < x < 8. Since f (0) > 0 and f (8) < 0, graphing techniques from calculus show that f has exactly 3 real roots. By (6.6.7), G = S5 . 4. f is irreducible by the rational root test. By the formula for the discriminant of a general cubic with a = 3, b = −2, c = 1, we have D = 9(−8) − 4(−8) − 27 − 18(6) = −175. Alternatively, if we replace X by X − a3 = X − 1, the resulting polynomial is g(X) = X 3 − 5X + 5, whose discriminant is −4(−5)3 − 27(25) = −175. In any event, D is not a square in Q, so G = S3 . (Notice also that g is irreducible by Eisenstein, so we could have avoided the rational root test at the beginning.) 5. If f is reducible, then it is the product of a linear factor and a quadratic polynomial g. If g is irreducible, then G is cyclic of order 2 (section 6.3, Problem 4). If g is reducible, then all roots of f are in the base ﬁeld, and G is trivial. 6. Let the roots be a, b + ic and b − ic. Then ∆ = (a − b − ic)(a − b + ic)2ic = ((a − b)2 + c2 )2ic and since i2 = −1, we have D < 0. Since D cannot be a square in Q, the Galois group is S3 . [This also follows from (6.6.7).] 7. If the roots are a, b and c, then D = (a − b)2 (a − c)2 (b − c)2 > 0. The result follows from (6.6.3). Section 6.7

√ 1. By (6.7.2), the √ Galois √ group √ √of Q( m)/Q is Z2 for m = 2, 3, 5, 7. It follows that the Galois group of Q( 2, 3, 5, 7)/Q is Z2 × Z2 × Z2 × Z2 . See (6.7.5), and note that Q contains a primitive square root of unity, namely -1. (It is not so easy to prove that the Galois group has order 16. One approach is via the texhnique of Section 7.3, Problems 9 and 10.) 2. Yes. Let E be the pth cyclotomic extension of Q, where p is prime. If p > 2, then Q does not contain a primitive pth root of unity. By (6.5.6), the Galois group is isomorphic to the group of units mod p, which is cyclic. 3. Since the derivative of f (X) = X n − a is nX n−1 = 0, it follows from (3.4.2) that f has n distinct roots β1 , . . . , βn in E. Since βin = a and βi−n = a−1 , there are n distinct nth roots of unity in E, namely 1 = β1 β1−1 , β2 β1−1 , . . . , βn β1−1 . Since the group of nth roots of unity is cyclic, there must be a primitive nth root of unity in E. 4. Each root of g is of the form ω i θ, so g0 = ω k θd for some k. Since ω p = 1, we have

page 5 of Solutions Chapters 6-10

g0p = θdp . But c = θp since θ is also a root of f , and the result follows. 5. By Problem 4 we have c = c1 = cad cbp = g0ap cbp = (g0a cb )p with g0a cb ∈ F . Thus g0a cb is a root of f in F . 6. [E : F (ω)] divides p and is less than p by (6.7.2); note that E is also a splitting ﬁeld for f over F (ω). Thus [E : F (ω)] must be 1, so E = F (ω). 7. F contains a primitive pth root of unity ω iﬀ E(= F (ω)) = F iﬀ X p − c splits over F . 8. By induction, σ j (θ) = θ + j, 0 ≤ j ≤ p − 1. Thus the subgroup of G that ﬁxes θ, hence ﬁxes F (θ), consists only of the identity. By the fundamental theorem, E = F (θ). 9. We have σ(θp − θ) = σ(θ)p − σ(θ) = (θ + 1)p − (θ + 1) = θp − θ in characteristic p. Thus θp − θ belongs to the ﬁxed ﬁeld of G, which is F . Let a = θp − θ, and the result follows. 10. Since f (θ) = 0, min(θ, F ) divides f . But the degree of the minimal polynomial is [F (θ) : F ] = [E : F ] = p = deg f . Thus f = min(θ, F ), which is irreducible. 11. Since θp −θ = a, we have (θ +1)p −(θ +1) = θp −θ = a. Inductively, θ, θ +1, . . . , θ +p−1 are distinct roots of f in E, and since f has degree p, we have found all the roots and f is separable. Since E is a splitting ﬁeld for f over F , we have E = F (θ). 12. By Problem 11, every root of f generates the same extension of F , namely E. But any monic irreducible factor of f is the minimal polynomial of at least one of the roots of f , and the result follows. 13. [E : F ] = [F (θ) : F ] = deg(min(θ, F )) = deg f = p. Thus the Galois group has prime order p and is therefore cyclic. Section 6.8 1. Take the real part of each term of the identity to get cos 3θ = cos3 θ + 3 cos θ(i sin θ)2 = cos3 θ − 3 cos θ(1 − cos2 θ); thus cos 3θ = 4 cos3 θ − 3 cos θ. If 3θ = π/3, we have cos π/3 = 1/2 = 4α3 − 3α so 8α3 − 6α − 1 = 0. But 8X 3 − 6X − 1 is irreducible over Q (rational root test), so α is algebraic over Q and [Q(α) : Q] = 3 (not a power √ of 2), a contradiction. √ 3 2. X 3 − 2 is irreducible by Eisenstein, so√[Q( 3 2) √ : Q] = 3 and 2 is not constructible. 3. The side of such a square would be π, so π, hence π, would be algebraic over Q, a contradiction. 4. ω is a root of X 2 − 2(cos 2π/n)X + 1 since cos 2π/n = 12 (ω + ω −1 ) and ω 2 − (ω + ω −1 )ω + 1 = 0. 5. By (6.5.2), (6.5.5) and (3.1.9), ϕ(n) = [Q(ω) : Q] = [Q(ω) : Q(cos 2π/n)][Q(cos 2π/n) : Q]. By Problem 4, [Q(ω) : Q(cos 2π/n)] = 2, and if the regular n-gon is constructible, then [Q(cos 2π/n) : Q] is a power of 2. The result follows. 6. By hypothesis, G =Gal(Q(ω)/Q) is a 2-group since its order is ϕ(n). Therefore every quotient group of G, in particular Gal(Q(cos 2π/n)/Q), is a 2-group. [Note that by (6.5.1), G is abelian, hence every subgroup of G is normal, and therefore every intermediate ﬁeld is a Galois extension of Q. Thus part 2c of the fundamental theorem (6.2.1) applies.] 7. By the fundamental theorem (speciﬁcally, by Section 6.2, Problem 4), there are ﬁelds Q = K0 ≤ K1 ≤ · · · ≤ Kr = Q(cos 2π/n) with [Ki : Ki−1 ] = 2 for all i = 1, . . . , r. Thus cos 2π/n is constructible. 8. If n = pe11 · · · perr , then (see Section 1.1, Problem 13) ϕ(n) = pe11 −1 (p1 − 1) · · · perr −1 (pr − 1).

page 6 of Solutions Chapters 6-10

If pi = 2, we must have ei = 1, and in addition, pi − 1 must be a power of 2. The result follows. 9. If m is not a power of 2, then m can be factored as ab where a is odd and 1 < b < m. In the quotient (X a + 1)/(X + 1), set X = 2b . It follows that (2m + 1)/(2b + 1) is an integer. Since 1 < 2b + 1 < 2m + 1, 2m + 1 cannot be prime. 10. Any permutation of the αi induces an automorphism of E which ﬁxes each ei , hence ﬁxes F . Thus the Galois group of f consists of all permutations of n letters. 11. Since Sn is not solvable, the general equation of degree n is not solvable by radicals if n ≥ 5. In other words,if n ≥ 5, there is no sequence of operations on e1 , . . . , en involving addition, subtraction, multiplication, division and extraction of mth roots, that will yield the roots of f . Section 6.9 1. If S is not maximal, keep adding elements to S until a maximal algebraically independent set is obtained. If we go all the way to T , then T is algebraically independent and spans E algebraically, hence is a transcendence basis. (Transﬁnite induction supplies the formal details.) 2. For the ﬁrst statement, take T = E in Problem 1. For the second statement, take S = ∅. 3. (i) implies (ii): Suppose that ti satisﬁes f (ti ) = b0 + b1 ti + · · · + bm tm i = 0, with bj ∈ F (T \ {ti }). By forming a common denominator for the bj , we may assume that the bj are polynomials in F [T \ {ti }] ⊆ F [T ]. By (i), bj = 0 for all j, so f = 0. (ii) implies (iii): Note that F (t1 , . . . , ti−1 ) ⊆ F (T \ {ti }). (iii) implies (i): Suppose that f is a nonzero polynomial in F [X1 , . . . , Xm ] such that r f (t1 , . . . , tm ) = 0, where m is as small as possible. Then f = h0 + h1 Xm + · · · + hr Xm r where the hj belong to F [X1 , . . . , Xm−1 ]. Now f (t1 , . . . , tm ) = b0 + b1 tm + · · · + br tm where bj = hj (t1 , . . . , tm−1 ). If the bj are not all zero, then tm is algebraic over F (t1 , . . . , tm−1 ), contradicting (iii). Thus bj ≡ 0, so by minimality of m, hj ≡ 0, so f = 0. 4. If S∪{t} is algebraically dependent over F , then there is a positive integer n and a nonzero polynomial f in F [X1 , . . . , Xn , Z] such that f (t1 , . . . , tn , t) = 0 for some t1 , . . . , tn ∈ S. Since S is algebraically independent over F , f must involve Z. We may write f = b0 + b1 Z + · · · + bm Z m where bm = 0 and the bj are polynomials in F [X1 , . . . , Xn ]. But then t is algebraic over F (S). Conversely, if t is algebraic over F (S), then for some positive integer n, there are elements t1 , . . . , tn ∈ S such that t is algebraic over F (t1 , . . . , tn ). By Problem 3, {t1 , . . . , tn , t} is algebraically dependent over F , hence so is S ∪ {t}. 5. Let A = {s1 , . . . , sm , t1 , . . . , tn } be an arbitrary ﬁnite subset of S ∪ T , with si ∈ S and tj ∈ T . By Problem 3, si is transcendental over F (s1 , . . . , si−1 ) and tj is transcendental over K(t1 , . . . , tj−1 ), hence over F (s1 , . . . , sm , t1 , . . . , tj−1 ) since S ⊆ K. Again by Problem 3, A is algebraically independent over F . Since A is arbitrary, S ∪ T is algebraically independent over F . Now if t ∈ K then {t} is algebraically dependent over K (t is a root of X − t). But if t also belongs to T , then T is algebraically dependent over K, contradicting the hypothesis. Thus K and T , hence S and T , are disjoint. 6. By Problem 5, S ∪ T is algebraically independent over F . By hypothesis, E is algebraic over K(T ) and K is algebraic over F (S). Since each t ∈ T is algebraic over F (S)(T ) = F (S ∪ T ), it follows that K(T ) is algebraic over F (S ∪ T ). By (3.3.5), E is algebraic over F (S ∪ T ). Therefore S ∪ T is a transcendence basis for E/F . 7. If T is algebraically independent over F , the map f (X1 , . . . , Xn ) → f (t1 , . . . , tn ) extends to an F -isomorphism of F (X1 , . . . , Xn ) and F (t1 , . . . , tn ). Conversely, assume that F (T ) is F -isomorphic to the rational function ﬁeld. By Problem 2, there is a transcendence basis B for F (T )/F such that B ⊆ T . By (6.9.7), the transcendence degree of F (T )/F is |T | = n. By (6.9.5) or (6.9.6), B = T , so T is algebraically independent over F . 8. The “if” part is clear since [K(z) : K] can’t be ﬁnite; if so, [F : K] < ∞. For the “only

page 7 of Solutions Chapters 6-10

if” part, z is algebraic over K(x), so let z n + ϕn−1 (x)z n−1 + · · · + ϕ0 (x) = 0, ϕi ∈ K(x). Clear denominators to get a polynomial f (z, x) = 0, with coeﬃcients of f in K. Now x must appear in f , otherwise z is not transcendental. Thus x is algebraic over K(z), so [K(z, x) : K(z)] < ∞. Therefore [F : K(z)] = [F : K(z, x)][K(z, x) : K(z)]. The ﬁrst term on the right is ﬁnite since K(x) ⊆ K(z, x), and the second term is ﬁnite, as we have just seen. Thus [F : K(z)] < ∞, and the result follows. ♣ 9. We have tr deg (C/Q) = c, the cardinality of C (or R). For if C has a countable transcendence basis z1 , z2 , . . . over Q, then C is algebraic over Q(z1 , z2 , . . .). Since a polynomial over Q can be identiﬁed with a ﬁnite sequence of rationals, it follows that |C| = |Q|, a contradiction. Section 7.1 1. Replace (iii) by (iv) and the proof goes through as before. If R is a ﬁeld, then in (iii) implies (i), x is an eigenvalue of C, so det(xI − C) = 0. 2. Replace (iii) by (v) and the proof goes through as before. [Since B is an A[x]-module, in (iii) implies (i) we have xβi ∈ B; when we obtain [det(xI − C)]b = 0 for every b ∈ B, the hypothesis that B is faithful yields det(xI − C) = 0.] 3. Multiply the equation by an−1 to get a−1 = −(cn−1 + · · · + c1 an−2 + c0 an−1 ) ∈ A. 4. Since A[b] is a subring of B, it is an integral domain. Thus if bz = 0 and b = 0, then z = 0. 5. Any linear transformation on a ﬁnite-dimensional vector space is injective iﬀ it is surjective. Thus if b ∈ B and b = 0, there is an element c ∈ A[b] ⊆ B such that bc = 1. Therefore B is a ﬁeld. 6. P is the preimage of Q under the inclusion map of A into B, so P is a prime ideal. The map a + P → a + Q is a well-deﬁned injection of A/P into B/Q, since P = Q ∩ A. Thus A/P can be viewed as a subring of B/Q. 7. If b + Q ∈ B/Q, then b satisﬁes an equation of the form xn + an−1 xn−1 + · · · + a1 x + a0 = 0, ai ∈ A. By Problem 6, b + Q satisﬁes the same equation with ai replaced by ai + P for all i. Thus B/Q is integral over A/P . 8. By Problems 3-5, A/P is a ﬁeld if and only if B/Q is a ﬁeld, and the result follows. (Note that since Q is a prime ideal, B/Q is an integral domain, as required in the hypothesis of the result just quoted.) Section 7.2

√ 1. By the quadratic formula, L = Q( b2 − 4c). Since b2 − 4c ∈ Q, we may write b2 − 4c = 2 s/t = st/t2 for relatively prime integers s and t. We also have s = uy 2 and where √ t = vz √ u, v, y, z ∈√Z, with u√and v relatively prime and square-free. Thus L = Q( uv) = Q( d). √ √ Q( e), then d = a + b e for rational numbers a and b. Thus d = 2. If Q( d) = √ √ a2 + b2 e + 2ab e, so e is√rational, a √ contradiction (unless a = 0 and b = 1). √ √ carry d into a + b e for rational numbers 3. Any isomorphism of Q( d) and Q( e) must √ a and b. Thus d is mapped to a2 + b2 e + 2ab e. But a Q-isomorphism maps d to d, and we

page 8 of Solutions Chapters 6-10

reach a contradiction as in Problem 2. 2 4. Since ωn = ω2n we have ωn ∈ Q(ω2n ), so Q(ωn ) ⊆ Q(ω2n ). If n is odd then n + 1 = 2r, so 2r 2 r ω2n = −ω2n = −(ω2n ) = −ωnr . Therefore Q(ω2n ) ⊆ Q(ωn ). 5. Let f be a monic polynomial over Z with f (x) = 0. If f is factorable over Q, then it is factorable over Z by (2.9.2). Thus min(x, Q) is the monic polynomial in Z[X] of least degree√such that f (x) = 0. √ 6. Q( −3) = Q(ω) where ω = − 12 + 12 −3 is a primitive cube root of unity. 7. If n = [L : Q], then an integral basis consists of n elements of L that are linearly independent over Z, hence over Q. (A linear dependence relation over Q can be converted to one over Z by multiplying by a common denominator.) Section 7.2

√ 1. By the quadratic formula, L = Q( b2 − 4c). Since b2 − 4c ∈ Q, we may write b2 − 4c = 2 s/t = st/t2 for relatively prime integers s and t. We also have s = uy 2 and where √ t = vz √ uv) = Q( d). u, v, y, z ∈√Z, with u√and v relatively prime and square-free. Thus L = Q( √ √ 2. If Q( d) = Q( e), then d = a + b e for rational numbers a and b. Thus d = √ √ a2 + b2 e + 2ab e, so e is√rational, a √ contradiction (unless a = 0 and b = 1). √ √ 3. Any isomorphism of Q( d) and Q( e) must carry d into a + b e for rational numbers √ a and b. Thus d is mapped to a2 + b2 e + 2ab e. But a Q-isomorphism maps d to d, and we reach a contradiction as in Problem 2. 2 4. Since ωn = ω2n we have ωn ∈ Q(ω2n ), so Q(ωn ) ⊆ Q(ω2n ). If n is odd then n + 1 = 2r, so 2r 2 r ω2n = −ω2n = −(ω2n ) = −ωnr . Therefore Q(ω2n ) ⊆ Q(ωn ). 5. Let f be a monic polynomial over Z with f (x) = 0. If f is factorable over Q, then it is factorable over Z by (2.9.2). Thus min(x, Q) is the monic polynomial in Z[X] of least degree√such that f (x) = 0. √ 6. Q( −3) = Q(ω) where ω = − 12 + 12 −3 is a primitive cube root of unity. 7. If n = [L : Q], then an integral basis consists of n elements of L that are linearly independent over Z, hence over Q. (A linear dependence relation over Q can be converted to one over Z by multiplying by a common denominator.) Section 7.4 1. If l(y) = 0, then (x, y) = 0 for all x. Since the bilinear form is nondegenerate, we must have y = 0. 2. Since V and V ∗ have the same dimension n, the map y → l(y) is surjective. 3. We have (xi , yj ) = l(yj )(xi ) = fj (xi ) = δij . Since the fj = l(yj ) form a basis, so do the yj . n 4. Write xi = k=1 aik yk , and take the inner product of both sides with xj to conclude that aij = (xi , xj ). 5. The “if” part was done in the proof of (7.4.10). If det C = ±1, then C −1 has coeﬃcients in Z by Cramer’s rule. √ 6. If √d ≡ 1 mod 4, then by (7.2.3), 1 and d form an integral basis. Since the trace of a + b d is 2a (Section 7.3, Problem 1), the ﬁeld discriminant is 2 0 D = det = 4d. 0 2d √ If d ≡ 1 mod 4, then 1 and 12 (1 + d) form an integral basis, and √ 1 1 d 1√ [ (1 + d)]2 = + + d. 2 4 4 2

page 9 of Solutions Chapters 6-10

Thus

2 D = det 1

1 d+1 2

= d.

7. The ﬁrst statement follows because multiplication of each element of a group G by a particular element g ∈ G permutes the elements of G. The plus and minus signs are balanced in P + N and P N , before and after permutation. We can work in a Galois extension of Q containing L, and each automorphism in the Galois group restricts to one of the σi on L. Thus P + N and P N belong to the ﬁxed ﬁeld of the Galois group, which is Q. 8. Since the xj are algebraic integers, so are the σi (xj ), as in the proof of (7.3.10). By (7.1.5), P and N , hence P + N and P N , are algebraic integers. By (7.1.7), Z is integrally closed, so by Problem 7, P + N and P N belong to Z. 9. D = (P − N )2 = (P + N )2 − 4P N ≡ (P + N )2 mod 4. But any square is congruent to 0 or 1 mod 4, and the nresult follows. 10. We have yi = j=1 aij xj with aij ∈ Z. By (7.4.3), D(y) = (det A)2 D(x). Since D(y) is square-free, det A = ±1, so A has an inverse with entries in Z. Thus x = A−1 y, as claimed. 11. Every algebraic integer is a Z-linear combination of the xi , hence of the yi by Problem 10. Since the yi form a basis for L over Q, they are linearly independent and the result follows. √ 12. No. For example, let L = Q( d), where d is a square-free integer with d ≡ 1 mod 4. (See Problem 6). The ﬁeld discriminant is 4d, which is not square-free. 13. This follows from the proof of (7.4.7). Section 7.5 1. A0 ⊂ A1 ⊂ A2 ⊂ · · · 2. Let a/pn ∈ B, where p does not divide a. There are integers r and s such that ra+spn = 1. Thus ra/pn = 1/pn in Q/Z, and An ⊆ B. If there is no upper bound on n, then 1/pn ∈ B for all n (note 1/pn = p/pn+1 = p2 /pn+2 , etc.), hence B = A. If there is a largest n, then for every m > n, B ∩ Am ⊆ An by maximality of n. Therefore B = An . 3. Let x1 , x2 , . . . be a basis for V . Let Mr be the subspace spanned by x1 , . . . , xr , and Lr the subspace spanned by the xj , j > r. If V is n-dimensional, then V = L0 > L1 > · · · > Ln−1 > Ln = 0 is a composition series since a one-dimensional subspace is a simple module. [V = Mn > Mn−1 > · · · > M1 > 0 is another composition series.] Thus V is Noetherian and Artinian. If V is inﬁnite-dimensional, then M1 < M2 < · · · violates the acc, and L0 > L1 > L2 > · · · violates the dcc. Thus V is neither Noetherian nor Artinian. [Note that if V has an uncountable basis, there is no problem; just take a countably inﬁnite subset of it.] 4. l(M ) is ﬁnite iﬀ M has a composition series iﬀ M is Noetherian and Artinian iﬀ N and M/N are Noetherian and Artinian iﬀ l(N ) and l(M/N ) are ﬁnite. 5. By Problem 4, the result holds when l(M ) = ∞, so assume l(M ), hence l(N ) and l(M/N ), ﬁnite. Let 0 < N1 < · · · < Nr = N be a composition series for N , and let N/N < (M1 + N )/N < · · · < (Ms + N )/N = M/N be a composition series for M/N . Then 0 < N 1 < · · · < N r < M1 + N < · · · < M s + N = M is a composition series for M . (The factors in the second part of the series are simple by the third isomorphism theorem.) It follows that l(M ) = r + s = l(N ) + l(M/N ). 6. By (7.5.9), R is a Noetherian S-module, hence a Noetherian R-module. (Any Rsubmodule T of R is, in particular, an S-submodule of R. Therefore T is ﬁnitely generated.) 7. Yes. Map a polynomial to its constant term and apply the ﬁrst isomorphism theorem to show that R ∼ = R[X]/(X). Thus R is a quotient of a Noetherian R-module, so is Noetherian by (7.5.7). 8. If there is an inﬁnite descending chain of submodules Mi of M , then the intersection N = ∩i Mi cannot be expressed as the intersection of ﬁnitely many Mi . By the correspondence theorem, ∩i (Mi /N ) = 0, but no ﬁnite intersection of the submodules Mi /N of M/N

page 10 of Solutions Chapters 6-10

is 0. Thus M/N is not ﬁnitely cogenerated. Conversely, suppose that M/N is not ﬁnitely cogenerated. By the correspondence theorem, we have ∩α Mα = N , but no ﬁnite intersection of the Mα is N . Pick any Mα and call it M1 . If M1 ⊆ Mα for all α, then M1 = N , a contradiction. Thus we can ﬁnd Mα = M2 such that M1 ⊃ M1 ∩ M2 . Continue inductively to produce an inﬁnite descending chain. Section 7.6 1. The “only if” part follows from (7.6.2). If the given condition is satisﬁed and ab ∈ P , then (a)(b) ⊆ P , hence (a) ⊆ P or (b) ⊆ P , and the result follows. 2. If xi ∈ / Pi for some i, then xi ∈ I \ ∪nj=1 Pj and we are ﬁnished. 3. Since I is an ideal, x ∈ I. Say x ∈ P1 . All terms in the sum that involve x1 belong to P1 by Problem 2. The remaining term x2 · · · xn is the diﬀerence of two elements in P1 , hence x2 · · · xn ∈ P1 . Since P1 is prime, xj ∈ P1 for some j = 1, contradicting the choice of xj . 4. The product of ideals is always contained in the intersection. If I and J are relatively prime, then 1 = x + y with x ∈ I and y ∈ J. If z ∈ I ∩ J, then z = z1 = zx + zy ∈ IJ. The general result follows by induction, along with the computation R = (I1 + I3 )(I2 + I3 ) ⊆ I1 I2 + I3 . Thus I1 I2 and I3 are relatively prime. 5. See (2.6.9). 6. Assume that R is not a ﬁeld, equivalently, {0} is not a maximal ideal. Thus by (7.6.9), every maximal idea is invertible. 7. Let r be a nonzero element of R such that rK ⊆ R, hence K ⊆ r−1 R ⊆ K. Thus K = r−1 R. Since r−2 ∈ K we have r−2 = r−1 s for some s ∈ R. But then r−1 = s ∈ R, so K ⊆ R and consequently K = R. 8. R = Rr = (P1 + P2 )r ⊆ P1r + P2 . Thus P1r and P2 are relatively prime for all r ≥ 1. Assuming inductively that P1r and P2s are relatively prime, we have P2s = P2s R = P2s (P1r + P2 ) ⊆ P1r + P2s+1 so R = P1r + P2s ⊆ P1r + (P1r + P2s+1 ) = P1r + P2s+1 completing the induction. Section 7.7

√ 1. By Section 7.3, Problem 1, the norms are 6,6,4 and 9. Now if x = a + b −5 and x = yz, then N (x) = a2 + 5b2 = N (y)N (z). The only algebraic integers of norm 1 are ±1, and there are no √ algebraic integers of norm 2 or 3. Thus there cannot be a nontrivial factorization of 1 ± −5, 2 √ or 3. √ 2. If (a + b −5)(c + d −5) = 1, take norms to get (a2 + 5b2 )(c2 + 5d2 ) = 1, so b = d = 0, a = ±1, c = ±1. 3. By Problem 2, if two factors are associates, then the quotient of the factors is ±1, which is impossible. 4. This is a nice application of the principle that divides means contains. The greatest common divisor is the smallest ideal containing both I and J, that is, I + J. The least common multiple is the largest ideal contained in both I and J, which is I ∩ J. 5. If I is a fractional ideal, then by (7.7.1) there is a fractional ideal I such that II = R. By deﬁnition of fractional ideal, there is a nonzero element r ∈ R such that rI is an integral ideal. If J = rI , then IJ = Rr, a principal ideal of R. √ 2 6. √ This is done just as in Problems 1-3, using the factorization 18 = (2)(3 ) = (1+ −17)(1− −17). √ √ 7. By (7.2.2), the algebraic integers are of the form a + b −3, a, b ∈ Z, or u2 + v2 −3 where

page 11 of Solutions Chapters 6-10

u and v are odd integers. If we require that the norm be 1, we only get ±1 in the ﬁrst case. But in the second case, we have u2 + 3v 2 = 4, so u = ±1, v = ±1. Thus if ω = ei2π/3 , the algebraic integers of norm 1 are ±1, ±ω, and ±ω 2 . Section 7.8 1. If Rx and Ry belong to P (R), then (Rx)(Ry)−1 = (Rx)(Ry −1 ) = Rxy −1 ∈ P (R), and the result follows from (1.1.2). 2. If C(R) is trivial, then every integral ideal I of R is a principal fractional ideal Rx, x ∈ K. But I ⊆ R, so x = 1x must belong to R, proving that R is a PID. The converse holds because every principal ideal is a√principal fractional√ideal. √ √ 3. 1 − −5 = 2 − (1 + −5) ∈ P2 , so (1 + −5)(1 − −5) = 6 ∈ P22 . 2 4. Since 2√∈ P2 , it follows P22 . √ that 4 ∈ P2 , so √ by Problem √ 3, 22 = 6 − 4 ∈ √ √ 5. (2, 1 + −5)(2, 1 + −5) = (4, 2(1 + −5), (1 + −5) ), and (1 + −5)2 = −4 + 2 −5. Thus each of the generators of the ideal P22 is divisible by 2, hence belongs to (2). Therefore P22 ⊆ (2). √ 6. x2 + 5 ≡ (x + 1)(x − 1) mod 3, which suggests that (3) = P3 P3 , where P3 = (3, 1 + −5) √ and P3 = (3, 1 − −5). √ √ 7. P3 P3 = (3, 3(1 − −5), 3(1 + −5), 6) ⊆ (3) since each generator of P3 P3 is divisible by 3. But 3 ∈ P3 ∩ P3 , hence 9 ∈ P3 P3 , and therefore 9 − 6 = 3 ∈ P3 P3 . Thus (3) ⊆ P3 P3 , and the result follows. Section 7.9 1. Using (1), the product is z = 4 + 2p + 4p2 + p3 + p4 . But 4 = 3 + 1 = 1 + p and 4p2 = p2 + 3p2 = p2 + p3 . Thus z = 1 + 3p + p2 + 2p3 + p4 = 1 + 2p2 + 2p3 + p4 . Using (2), we are multiplying x = {2, 5, 14, 14, . . .} by y = {2, 2, 11, 11, . . .}. Thus z0 = 4, z1 = 10, z2 = 154, z3 = 154, z4 = 154, and so on. But 4 ≡ 1 mod 3, 10 ≡ 1 mod 9, 154 ≡ 19 mod 27, 154 ≡ 73 mod 81, 154 ≡ 154 mod 243. The standard form is {1, 1, 19, 73, 154, 154, . . .}. As a check, the product is (2+3+9)(2+9)=154, whose base 3 expansion is 1 + 0(3) + 2(9) + 2(27) + 1(81) as found above. 2. We have a0 = −1 and an = 0 for n ≥ 1; equivalently, xn = −1 for all n. In standard form, x0 = p − 1, x1 = p2 − 1, x2 = p3 − 1, . . . .Since (pr − 1) − (pr−1 − 1) = (p − 1)(pr−1 ), the series representation is (p − 1) + (p − 1)p + (p − 1)p2 + · · · + (p − 1)pn + · · · . The result can also be obtained by multiplying by -1 on each side of the equation 1 = (1 − p)(1 + p + p2 + · · ·). 3. Let x be a nonzero element of GF (q). By (6.4.1), xq−1 = 1, so |x|q−1 = 1. Thus |x| is a root of unity, and since absolute values are nonnegative real, we must have |x| = 1, and the result follows. 4. If the absolute value is nonarchimedian, then S is bounded by (7.9.6). If the absolute value is archimedian, then by (7.9.6), |n| > 1 for some n. But then |nk | = |n|k → ∞ as k → ∞. Therefore S is unbounded. 5. A ﬁeld of prime characteristic p has only ﬁnitely many integers 0, 1, . . . , p − 1. Thus the set S of Problem 4 must be bounded, so the absolute value is nonarchimedian. 6. The “only if” part is handled just as in calculus. For the “if” part, note that by (iv) of (7.9.5), we have |zm + zm+1 + · · · + zn | ≤ max{|zi | : m ≤ i ≤ n} → 0 as m, n → ∞. Thus the nth partial sums form a Cauchy sequence, which must converge to an element in Qp . 7. Since n! = 1 · 2 · · · p · · · 2p · · · 3p · · ·, it follows from (7.9.2) and (7.9.3) that if rp ≤ n < (r + 1)p, then |n!| = 1/pr . Thus |n!| → 0 as n → ∞. 8. No. Although |pr | = 1/pr → 0 as r → ∞, all integers n such that rp < n < (r + 1)p

page 12 of Solutions Chapters 6-10

have p-adic absolute value 1, by (7.9.2). Thus the sequence of absolute values |n| cannot converge, hence the sequence itself cannot converge. Section 8.1 1. If x ∈ V and f1 (x) = 0, then f2 (x) must be 0 since f1 f2 ∈ I(V ); the result follows. 2. By Problem 1, V ⊆ V (f1 ) ∪ V (f2 ). Thus V = (V ∩ V (f1 )) ∪ (V ∩ V (f2 )) = V1 ∪ V2 . Since f1 ∈ / I(V ), there exists x ∈ V such that f1 (x) = 0. Thus x ∈ / V1 , so V1 ⊂ V ; similarly, V2 ⊂ V . 3. I(V ) ⊇ I(W ) by (4). If I(V ) = I(W ), let V = V (S), W = V (T ). Then IV (S) = IV (T ), and by applying V to both sides, we have V = W by (6). 4. Let x ∈ V ; if f1 (x) = 0, then since f1 ∈ I(V1 ), we have x ∈ / V1 . But then x ∈ V2 , and therefore f2 (x) = 0 (since f2 ∈ I(V2 )). Thus f1 f2 = 0 on V , so f1 f2 ∈ I(V ). 5. If V is reducible, then V is the union of proper subvarieties V1 and V2 . If V1 is reducible, then it too is the union of proper subvarieties. This decomposition process must terminate in a ﬁnite number of steps, for otherwise by Problems 1-4, there would be a strictly increasing inﬁnite sequence of ideals, contradictingthe fact that k[X1 , . . . , Xn ] is Noetherian. 6. If V = i Vi = j Wj , then Vi = j (Vi Wj ), so by irreducibility, Vi = Vi Wj for some j. Thus Vi ⊆ Wj , and similarly Wj ⊆ Vk for some k. But then Vi ⊆ Vk , hence i = k (otherwise we would have discarded Vi ). Thus each Vi can be paired with a corresponding Wj , and vice versa. 7. By hypothesis, An = ∪(An \ V (Ii )). Taking complements, we have ∩V (Ii ) = ∅. But by (8.1.2), ∩V (Ii ) = V (∪Ii ) = V (I), so by the weak Nullstellensatz, I = k[X1 , . . . , Xn ]. Thus the constant polynomial 1 belongs to I. 8. Suppose that the open sets An \ V (Ii ) cover An . By Problem 7, 1 ∈ I, hence 1 belongs to a ﬁnite sum i∈F Ii . Since 1 never vanishes, V ( i∈F Ii ) = ∅. By (8.1.2), ∩i∈F Vi = ∅, where Vi = V (Ii ). Taking complements, we have ∪i∈F (An \ Vi ) = An . Thus the original open covering of An has a ﬁnite subcovering, proving compactness. Section 8.2 1. If a ∈ / (a1 , . . . , ak ), then g or some other element of I would extend the inductive process to step k + 1. 2. In going from di to di+1 we are taking the minimum of a smaller set. 3. By minimality of m, a ∈ / (a1 , . . . , am−1 ), hence fm and g satisfy conditions 1 and 2. By choice of fm we have dm ≤ d. (If m = 1, then d1 ≤ d by choice of f1 .) 4. Let f be the unique ring homomorphism from R[X1 , . . . , Xn ] to S such that f is the identity on R and f (Xi ) = xi , i = 1, . . . , n. (For example, if a ∈ R, then aX12 X47 → ax21 x74 .) Since the image of f contains R and {x1 , . . . , xn }, f is surjective and the result follows. 5. By the Hilbert basis theorem, R[X1 , . . . , Xn ] is a Noetherian ring, hence a Noetherian R[X1 , . . . , Xn ]-module. By (7.5.7), S is a Noetherian R[X1 , . . . , Xn ]-module. But the submodules of S considered as an R[X1 , . . . , Xn ]-module coincide with the submodules of S as an S-module. (See Section 4.2, Problems 6 and 7; note that the kernel of the homomorphism f of Problem 4 annihilates S.) Thus S is a Noetherian S-module, that is, a Noetherian ring. Section 8.3 1. Suppose that xy ∈ J with x ∈ / J and y ∈ / J. By maximality of J, the ideal J + (x) contains an element s ∈ S. Similarly, J + (y) contains an element t ∈ S. But then st ∈ (J + (x))(J + (y)) ⊆ J + (xy) ⊆ J, so S ∩ J = ∅, a contradiction. √ 2. Let S = {1, f, f 2 , . . . , f r , . . .}. Then I ∩S = ∅ since f ∈ / I. By Problem 1, I is contained in a prime ideal P disjoint from S. But f ∈ S, so f cannot belong to P , and the result follows. 3. The “if” part follows because f and f r have the same zero-set. Conversely, if V (f ) =

page 13 of Solutions Chapters 6-10

V (g), then by the Nullstellensatz, (f )= (g), and the result follows. 4. W ⊆ V since (t4 )2 = (t3 )(t5 ) and (t5 )2 = (t3 )2 (t4 ); L ⊆ V by direct veriﬁcation. Conversely, if y 2 = xz and z 2 = x2 y, let t = y/x. (If x = 0, then y = z = 0 and we can take t = 0.) Then z = y 2 /x = (y/x)2 x = t2 x, and z 2 = x2 y. Therefore z 2 = t4 x2 = x2 y, hence y = t4 . If t = 0 then y = 0, hence z = 0 and (x, y, z) ∈ L. Thus assume t = 0. But then x = y/t = t3 and z = t2 x = t5 . 5. We will show that I(V ) is a prime ideal (see the exercises in Section 8.1). If f g ∈ I(V ), then f g vanishes on V . Using the parametric form, we have f (t, t2 , t3 )g(t, t2 , t3 ) = 0 for all complex numbers t. Since we are now dealing with polynomials in only one variable, either f (t, t2 , t3 ) = 0 for all t or g(t, t2 , t3 ) = 0 for all t. Thus f ∈ I(V ) or g ∈ I(V ). 6. (a) x = 2t/(t2 + 1), y = (t2 − 1)/(t2 + 1) (b) x = t2 , y = t3 (c) x = t2 − 1, y = t(t2 − 1) 7. (Following Shafarevich, Basic Algebraic Geometry, Vol.1, page 2.) We can assume that x appears in f with positive degree. Viewing f and g as polynomials in k(y)[x], a PID, f is still irreducible because irreducibility over an integral domain implies irreducibility over the quotient ﬁeld. If g = f h where h is a polynomial in x with coeﬃcients in k(y), then by clearing denominators we see that f must divide g in k[x, y], a contradiction. (Since f is irreducible, it must either divide g or a polynomial in y alone, and the latter is impossible because x appears in f .) Thus f does not divide g in k(y)[x]. Since f and g are relatively prime, there exist s, t ∈ k(y)[x] such that f s + gt = 1. Clearing denominators, we get u, v ∈ k[x, y] such that f u + gv = a, where a is a nonzero polynomial in y alone. Now if α, β ∈ k and f (α, β) = g(α, β) = 0, then a(β) = 0, and this can only happen for ﬁnitely many β. For any ﬁxed β, consider f (x, β) = 0. If this polynomial in x is not identically 0, then there are only ﬁnitely many α such that f (α, β) = 0, and we are ﬁnished. Thus assume f (x, β) ≡ 0. Then f (x, y) = f (x, y) − f (x, β) = (y − β)h in k(x)[y], contradicting the irreducibility of f . Section 8.4 1. Since f = 0 iﬀ some fi = 0, V (f ) is the union of the V (fi ). Since each fi is irreducible, the ideal Ii = (fi ) is prime by (2.6.1), hence V (Ii ) = V (fi ) is an irreducible subvariety of V (f ). [See the problems in Section 8.1, along with the Nullstellensatz and the fact that every prime ideal is a radical ideal (Section 8.3, Problem 2).] No other decomposition is possible, for if V (fi ) ⊆ V (fj ), then (fi ) ⊇ (fj ). This is impossible if fi and fj are distinct irreducible factors of f . 2. By the Nullstellensatz, IV (f ) = (f ), and we claim that (f ) = (f1 · · · fr ). For if g ∈ (f1 · · · fr ), then a suﬃciently high power of g will belong to (f). Conversely, if g m = hf , then each fi divides g m , and since the fi are irreducible, each fi divides g, so (f1 · · · fr ) divides g. 3. By Problem 1, f is irreducible if and only if V (f ) is an irreducible hypersurface. If f and g are irreducible and V (f ) = V (g), then as in Problem 1, (f ) = (g), so f = cg for some nonzero constant c (Section 2.1, Problem 2). Thus f → V (f ) is a bijection between irreducible polynomials and irreducible hypersurfaces, if the polynomials f and cf, c = 0, are identiﬁed. 4. This follows from the deﬁnition of I(X) in (8.1.3), and the observation that a function vanishes on a union of sets iﬀ it vanishes on each of the sets. 5. By Section 8.1, Problem 5, every variety V is the union of ﬁnitely many irreducible subvarieties V1 , . . . , Vr . By Problem 4, I(V ) = ∩ri=1 I(Vi ). By the Problems in Section 8.1, each I(Vi ) is a prime ideal. By (8.4.3), every radical ideal is I(V ) for some variety V , and the result follows. 6. By Section 8.1, Problem 6, and the inclusion-reversing property of I (part (4) of (8.1.3)), the decomposition is unique if we discard any prime ideal that properly contains another one. In other words, we retain only the minimal prime ideals. 7. If f is any irreducible factor of any of the fi , then f does not divide g. Thus for some

page 14 of Solutions Chapters 6-10

j = i, f does not divide fj . By Problem 7 of Section 8.3, the simultaneous equations f = fj = 0 have only ﬁnitely many solutions, and consequently X is a ﬁnite set. 8. With notation as in Problem 7, fi = ghi , where the gcd of the hi is constant. Thus X is the union of the algebraic curve deﬁned by g = 0 and the ﬁnite set deﬁned by h1 = · · · = hm = 0. (This analysis does not apply when X is deﬁned by the zero polynomial, in which case X = A2 .) 9. If k = R, the zero-set of x2 + y 2n is {(0, 0)} for all n = 1, 2, . . .. If k is algebraically closed, then as a consequence of the Nullstellensatz, V (f ) = V (g) with f and g irreducible implies that f = cg for some constant c. (See Problem 3). 10. Let k = F2 , and let I be the ideal of k[X] generated by f (X) = X 2 + X + 1. Since f is irreducible, I is a maximal ideal (Section 3.1, Problem 8), in particular, I is proper. But f (0) and f (1) are nonzero, so V (I) is empty, contradicting the weak Nullstellensatz. Section 8.5 1. If x ∈ / M , then the ideal generated by M and x is R, by maximality of M . Thus there exists y ∈ M and z ∈ R such that y + zx = 1. By hypothesis, zx, hence x, is a unit. Take the contrapositive to conclude that every nonunit belongs to M . 2. Any additive subgroup of the cyclic additive group of Zpn must consist of multiples of some power of p, and it follows that every ideal is contained in (p), which must therefore be the unique maximal ideal. 3. No. A can be nilpotent, that is, some power of A can be 0. The set will be multiplicative if A is invertible. 4. S −1 (gf ) takes m/s to g(f (m))/s, as does S −1 gS −1 f . If f is the identity on M , then S −1 f is the identity on S −1 M . 5. By hypothesis, gf = 0, so S −1 gS −1 f = S −1 gf = S −1 0 = 0. Thus im S −1 f ⊆ ker S −1 g. Conversely, let x ∈ N, s ∈ S, with x/s ∈ ker S −1 g. Then g(x)/s = 0/1, so for some t ∈ S we have tg(x) = g(tx) = 0. Therefore tx ∈ ker g = im f , so tx = f (y) for some y ∈ M . We now have x/s = f (y)/st = (S −1 f )(y/st) ∈ im S −1 f . 6. The set of nonuits is M = {f /g : g(a) = 0, f (a) = 0}, which is an ideal. By (8.5.9), R is a local ring with maximal ideal M . 7. The sequence 0 → N → M → M/N → 0 is exact, so by Problem 5, 0 → NS → MS → (M/N )S → 0 is exact. (If f is one of the maps in the ﬁrst sequence, the corresponding map in the second sequence is S −1 f .) It follows from the deﬁnition of localization of a module that NS ≤ MS , and by exactness of the second sequence we have (M/N )S ∼ = MS /NS , as desired. Section 8.6

√ √ 1. If xm belongs to the√intersection of the Ii , then x belongs to each I i , so x ∈ ∩ni=1 I i . n mi m n Conversely, n if x ∈ ∩i=1 I i , let x ∈ Ii . If m is the maximum of the mi , then x ∈ ∩i=1 Ii , so x ∈ ∩i=1 Ii . 2. We are essentially setting X = Z = 0 in R, and this collapses R down to k[Y ]. Formally, map f + I to g + I, where g consists of those terms in f that do not involve X or Z. Then R/P ∼ = k[Y ], an integral domain. Therefore P is prime. √ 3. (X + I)(Y + I) = Z 2 + I ∈ P 2 , but X + I ∈ / P 2 and Y + I ∈ / P2 = P. 4. P1 is prime because R/P1 ∼ = k[Y ], an integral domain. P2 is maximal by (8.3.1), so P22 is P2 -primary by (8.6.6). The radical of Q is P2 , so by (8.6.5), Q is P2 -primary. 5. The ﬁrst assertion is that (X 2 , XY ) = (X) ∩ (X, Y )2 = (X) ∩ (X 2 , XY, Y 2 ) and the second is (X 2 , XY ) = (X) ∩ (X 2 , Y ). In each case, the left side is contained in the right side by deﬁnition of the ideals involved. The inclusion from right to left follows because if f (X, Y )X = g(X, Y )Y 2 (or f (X, Y )X =

page 15 of Solutions Chapters 6-10

g(X, Y )Y ), then g(X, Y ) must involve X and f (X, Y ) must involve Y . Thus f (X, Y )X is a polynomial multiple of XY . 6. By (8.6.9), a proper ideal I can be expressed as the intersection of ﬁnitely many primary ideals Qi . If Qi is Pi -primary, then by Problem 1, √ I = I = ∩i Qi = ∩i Pi . √ √ 7. Since X 3 and Y n belong√to In , we have X, Y ∈ In , so (X, Y ) ⊆ In . By (8.3.1), √ (X, Y ) is a maximal ideal. Since In is proper (it does not contain 1), we have (X, Y ) = In . By (8.6.5), In is primary. Section 8.7 1. [(x, y + y ) + G] − [(x, y) + G] − [(x, y ) + G] = 0 since (x, y + y ) − (x, y) − (x, y ) ∈ G; r[(x, y) + G] − [(rx, y) + G] = 0 since r(x, y) − (rx, y) ∈ G; the other cases are similar. 2. Let a and b be integers such that am + bn = 1. If x ∈ Zm and y ∈ Zn , then x ⊗ y = 1(x ⊗ y) = a(mx ⊗ y) + b(x ⊗ ny) = 0 since z ⊗ 0 = 0 ⊗ z = 0. 3. Let A be a torsion abelian group, that is, each element of A has ﬁnite order. If x ∈ A and y ∈ Q, then nx = 0 for some positive integer n. Thus x ⊗ y = n(x ⊗ (y/n)) = nx ⊗ (y/n) = 0 ⊗ (y/n) = 0. 4. We have h = g h = g gh and h = gh = gg h . But if P = T and f = h, then g = 1T makes the diagram commute, as does g g. By the uniqueness requirement in the universal mapping property, we must have g g = 1T , and similarly gg = 1T . Thus T and T are isomorphic. 5. n ⊗ x = n(1 ⊗ x) = 1 ⊗ nx = 1 ⊗ 0 = 0. 6. nZ ⊗ Zn ∼ = Z ⊗ Zn ∼ = Zn by (8.7.6), with n ⊗ x → 1 ⊗ x → x, and since x = 0, n ⊗ x cannot be 0. 7. We have a bilinear map (f, g) → f ⊗ g from HomR (M, M )× HomR (N, N ) to HomR (M ⊗R N, M ⊗R N ), and the result follows from the universal mapping property of tensor products. 8. In terms of matrices, we are to prove that Mm (R) ⊗ Mn (R) ∼ = Mmn (R). This follows because Mt (R) is a free R-module of rank t2 . Section 8.8 1. y1 · · · (yi + yj ) · · · (yi + yj ) · · · yp = y1 · · · yi · · · yi · · · yp + y1 · · · yi · · · yj · · · yp + y1 · · · yj · · · yi · · · yp + y1 · · · yj · · · yj · · · yp . The left side, as well as the ﬁrst and last terms on the right, are zero by deﬁnition of N . Thus y1 · · · yi · · · yj · · · yp = −y1 · · · yj · · · yi · · · yp , as asserted. 2. If π is any permutation of {a, . . . , b}, then by Problem 1, xπ(a) · · · xπ(b) = (sgn π)xa · · · xb . The left side will be ±xa · · · xb , regardless of the particular permutation π, and the result follows. 3. The multilinear map f induces a unique h : M ⊗p → Q such that h(y1 ⊗ · · · ⊗ yp ) = f (y1 , . . . , yp ). Since f is alternating, the kernel of h contains N , so the existence and uniqueness of the map g follows from the factor theorem. 4. By Problem 3, there is a unique R-homomorphism g : Λn M → R such that g(y1 · · · yn ) = f (y1 , . . . , yn ). In particular, g(x1 · · · xn ) = f (x1 , . . . , xn ) = 1 = 0. Thus x1 · · · xn = 0. If r is any nonzero element of R, then g(rx1 x2 · · · xn ) = f (rx1 , . . . , xn ) = r, so rx1 · · · xn = 0. By Problem 2, {x1 · · · xn } is a basis for Λn M . 5. Fix the set of indices I0 and its complementary set J0 . If I aI xI = 0, xI ∈ R, multiply both sides on the right by xJ0 . If I = I0 , then xI xJ0 = 0 by deﬁnition of N . Thus aI0 xI0 xJ0 = ±aI0 x1 · · · xn = 0. By Problem 4, aI0 = 0. Since I0 is arbitrary, the result follows.

page 16 of Solutions Chapters 6-10

6. We have R0 ⊆ S by deﬁnition of S. Assume that Rm ⊆ S for m = 0, 1, . . . , n − 1, and r let a ∈ Rn (n > 0). Then a ∈ I, so a = i=1 ci xi where (since xi ∈ Rni and R is the direct sum of the Rm ) ci ∈ Rn−ni . By induction hypothesis, ci ∈ S, and since xi ∈ S by deﬁnition of S, we have a ∈ S, completing the induction. 7. The “if” part follows from Section 8.2, Problem 5, so assume R Noetherian. Since R0 ∼ = R/I, it follows that R0 is Noetherian. Since R is Noetherian, I is ﬁnitely generated, so by Problem 6, R = S, a ﬁnitely generated R0 -algebra. Section 9.1 1. Assume R is simple, and let x ∈ R, x = 0. Then Rx coincides with R, so 1 ∈ Rx. Thus there is an element y ∈ R such that yx = 1. Similarly, there is an element z ∈ R such that zy = 1. Therefore z = z1 = zyx = 1x = x, so xy = zy = 1 and y is a two-sided inverse of x. Conversely, assume that R is a division ring, and let x be a nonzero element of the left ideal I. If y is the inverse of x, then 1 = yx ∈ I, so I = R and R is simple. 2. I is proper because f (1) = x = 0, and R/I ∼ = Rx by the ﬁrst isomorphism theorem. 3. The “if” part follows from the correspondence theorem, so assume that M is simple. If x is a nonzero element of M , then M = Rx by simplicity. If I = ker f as in Problem 2, then M∼ = R/I, and I is maximal by the correspondence theorem. 4. The “only if” part was done in Problem 3, so assume that M is not simple. Let N be a submodule of M with 0 < N < M . If x is a nonzero element of N , then Rx ≤ N < M , so x cannot generate M . 5. By Problem 3, a simple Z-module is isomorphic to Z/I, where I is a maximal ideal of Z. By Section 2.4, Problems 1 and 2, I = (p) where p is prime. 6. As in Problem 5, a simple F [X]-module is isomorphic to F [X]/(f ), where f is an irreducible polynomial in F [X]. (See Section 3.1, Problem 8.) 7. If x is a nonzero element of V and y an arbitrary element of V , there is an endomorphism f such that f (x) = y. Therefore V = (Endk V )x. By Problem 4, V is a simple Endk (V )module. 8. By (4.7.4), every such short exact sequence splits iﬀ for any submodule N ≤ M , M ∼ = N ⊕ P , where the map N → M can be identiﬁed with inclusion and the map M → P can be identiﬁed with projection. In other words, every submodule of M is a direct summand. Equivalently, by (9.1.2), M is semisimple. Section 9.2 1. Unfortunately, multiplication by r is not necessarily an R-endomorphism of M , since r(sx) = (rs)x, which need not equal s(rx) = (sr)x. 2. Let x be a generator of M , and deﬁne f : R → M by f (r) = rx. By the ﬁrst isomorphism theorem, M ∼ = R/ann M . The result follows from the correspondence theorem. 3. Let M = Zp ⊕ Zp where p is prime. Then M is not a simple Z-module, but ann M = pZ is a maximal ideal of Z. 4. The computation given in the statement of the problem shows that (1,0) is a generator of V , hence V is cyclic. But N = {(0, b) : b ∈ F } is a nontrivial proper submodule of V . (Note that T (0, b) = (0, 0) ∈ N .) Therefore V is not simple. 5. Since F is algebraically closed, f has an eigenvalue λ ∈ F . Thus the kernel of f − λI is not zero, so it must be all of M . Therefore f (m) = λm for all m ∈ M . 6. If r ∈ I and s + I ∈ R/I, then r(s + I) = rs + I. But if I is not a right ideal, we cannot guarantee that rs belongs to I. Section 9.3 1. For each j = 1, . . . , n, there is a ﬁnite subset I(j) of I such that

xj belongs to the direct sum of the Mi , i ∈ I(j). If J is the union of the I(j), then M ⊆ i∈J Mi ⊆ M , so M is the

page 17 of Solutions Chapters 6-10

direct sum of the Mi , i ∈ J. 2. Each simple module Mi , i = 1, . . . , n, is cyclic (Section 9.1, Problem 4), and therefore can be generated by a single element xi . Thus M is generated by x1 , . . . , xn . 3. A left ideal is simple iﬀ it is minimal, so the result follows from (9.1.2). 4. No. If it were, then by Section 9.1, Problem 5, Z would be a direct sum of cyclic groups of prime order. Thus each element of Z would have ﬁnite order, a contradiction. 5. By (4.6.4), every ﬁnite abelian group is the direct sum of various Zp , p prime. If p and q are distinct primes, then Zp ⊕ Zq ∼ = Zpq by the Chinese remainder theorem. Thus Zn can be assembled from cyclic groups of prime order as long as no prime appears more than once in the factorization of n. (If Zp ⊕ Zp is part of the decomposition, the group cannot be cyclic.) Consequently, Zn is semisimple if and only if n is square-free. 6. This follows from Section 9.1, Problem 8. (In the ﬁrst case, B is semisimple by hypothesis, and in the second case A is semisimple. The degenerate case M = 0 can be handled directly.) 7. Conditions (a) and (b) are equivalent by (9.3.2) and the deﬁnition of semisimple ring. By Problem 6, (b) implies both (c) and (d). To show that (c) implies (b) and (d) implies (b), let M be a nonzero R-module, with N a submodule of M . By hypothesis, the sequence 0 → N → M → M/N → 0 splits. (By hypothesis, M/N is projective in the ﬁrst case and N is injective in the second case.) By Section 9.1, Problem 8, M is semisimple. Section 9.4 1. If there is an inﬁnite descending sequence I1 ⊃ I2 ⊃ · · · of left ideals, we can proceed exactly as in (9.4.7) to reach a contradiction. 2. Let I1 be any nonzero left ideal. If I1 is simple, we are ﬁnished. If not, there is a nonzero left ideal I2 such that I1 ⊃ I2 . If we continue inductively, the Artinian hypothesis implies that the process must terminate in a simple left ideal It . 3. By Problem 2, the ring R has a simple R-module M . The hypothesis that R has no nontrivial two-sided ideals implies that we can proceed exactly as in (9.4.6) to show that M is faithful. 4. If V is inﬁnite-dimensional over D, then exactly as in (9.4.7), we ﬁnd an inﬁnite descending chain of left ideals, contradicting the assumption that R is Artinian. 5. By Problem 4, V is a ﬁnite-dimensional vector space over D, so we can reproduce the discussion preceding (9.4.7) to show that R ∼ = EndD (V ) ∼ = Mn (Do ). 6. The following is a composition series: 0 < M1 < M1 ⊕ M2 < · · · < M1 ⊕ M2 ⊕ · · · ⊕ Mn = M. 7. By (9.1.2), M is a direct sum of simple modules. If the direct sum is inﬁnite, then we can proceed as in Problem 6 to construct an inﬁnite ascending (or descending) chain of submodules of M , contradicting the hypothesis that M is Artinian and Noetherian. Section 9.5 1. If g ∈ ker ρ, then gv = v for every v ∈ V . Take v = 1G to get g1G = 1G , so g = 1G and ρ is injective. 2. (gh)(v(i)) = v(g(h(i))) and g(h(v(i))) = g(v(h(i))) = v(g(h(i))). Also, 1G (v(i)) = v(1G (i)) = v(i). 3. We have g(v(1)) = v(4), g(v(2)) = v(2), g(v(3)) = v(1), g(v(4)) = v(3), so   0 0 1 0 0 1 0 0 [g] =  . 0 0 0 1 1 0 0 0 4. We have gv1 = v2 , gv2 = v3 = −v1 − v2 ; hv1 = v1 , hv2 = v3 = −v1 − v2 . Thus 0 −1 1 −1 [g] = , h= . 1 −1 0 −1

page 18 of Solutions Chapters 6-10

5. We have v1 = e1 and v2 = − 12 e1 +  P −1 =  and

1 2

√

3e2 . Thus

1

− 12

0

1 2

√





, P = 

1 0

3

√  3  √ 2 3 3 1 3

√  − 12 3 , [g] = P −1 [g]P =  √ 1 1 −2 2 3 1 0 −1 [h] = P [h]P = . 0 −1 

− 12

6. Check by direct computation that the matrices A and B satisfy the deﬁning relations of D8 : A4 = I, B 2 = I, AB = BA−1 . (See Section 5.8.) 7. Yes. Again, check by direct computation that the matrices Ai B j , i = 0, 1, 2, 3, j = 0, 1, are distinct. Thus if g ∈ D8 and ρ(g) = I, then g is the identity element of D8 . Section 9.6 1. Let W be the one-dimensional subspace spanned by v1 + v2 + v3 . Since any permutation in S3 permutes the vi , v ∈W implies gv ∈ W . 1 0 r 2. Multiplying [a ] by and , we have ar v1 = v1 and ar v2 = rv1 + v2 . Since W is 0 1 spanned by v1 , it is closed under the action of G and is therefore a kG-submodule. 3. If x x [ar ] =c y y then x + ry = cx and y = cy. If y = 0 then c = 1, so ry, hence y, must be 0. Thus c = 1 and x is arbitrary, so that any one-dimensional kG-submodule must coincide with W . 4. If V = W ⊕ U , where U is a kG-submodule, then U must be one-dimensional. By Problem 3, W = U , and since W = 0, this is impossible. 5. If M is semisimple and either Noetherian or Artinian, then M is the direct sum of ﬁnitely many simple modules. These simple modules are the factors of a composition series, and the result follows from (7.5.12). 6. Let e be the natural projection on M1 . Then e is an idempotent, e = 0 since M1 = 0, and e = 1 since e = 0 on M2 and M2 = 0. 7. Let e be a nontrivial idempotent, and deﬁne e1 = e, e2 = 1 − e. By direct computation, e1 and e2 are nontrivial idempotents that are orthogonal. Take M1 = e1 (M ), M2 = e2 (M ). Then M1 and M2 are nonzero submodules with M = M1 + M2 . To show that the sum is direct, let z = e1 x = e2 y ∈ M1 ∩ M2 , with x, y ∈ M . Then e1 z = e1 e2 y = 0, and similarly e2 z = 0. Thus z = 1z = e1 z + e2 z = 0. Section 9.7 1. If M = Rx, deﬁne f : R → M by f (r) = rx. By the ﬁrst isomorphism theorem, M ∼ = R/ker f . Moreover, ker f = ann(M ). Conversely, R/I is cyclic since it is generated by 1 + I. 2. If N is a maximal submodule of M , then N is the kernel of the canonical map of M onto the simple module M/N . Conversely, if f : M → S, S simple, then f (M ) is 0 or S, so f = 0 or S ∼ = M/ker f . Thus ker f is either M or a maximal submodule of M . The intersection of all kernels therefore coincides with the intersection of all maximal submodules. [If there are no maximal submodules, then the intersection of all kernels is M .] 3. By the correspondence theorem, the intersection of all maximal left ideals of R/I is 0. This follows because the intersection of all maximal left ideals of R containing I is the intersection of all maximal left ideals of R. [Note that J(R) is contained in every maximal

page 19 of Solutions Chapters 6-10

left ideal, and I = J(R).] 4. Let g be an R-module homomorphism from N to the simple R-module S. Then gf : M → S, so by Problem 2, J(M ) ⊆ ker (gf ). But then f (J(M )) ⊆ ker g. Take the intersection over all g to get f (J(M )) ⊆ J(N ). 5. Suppose a ∈ J(R). If 1 + ab is a nonunit, then it belongs to some maximal ideal M . But then a belongs to M as well, and therefore so does ab. Thus 1 ∈ M , a contradiction. Now assume a ∈ / J(R), so that for some maximal ideal M , a ∈ / M . By maximality, M + Ra = R, so 1 = x + ra for some x ∈ M and r ∈ R. Since x belongs to M , it cannot be a unit, so if we set b = −r, it follows that 1 + ab is a nonunit. 6. By the correspondence theorem, there is a bijection, given by ψ(A) = A/N , between maximal submodules of M containing N and maximal submodules of M/N . Since N ≤ J(M ) by hypothesis, a maximal submodule of M containing N is the same thing as a maximal submodule of M . Thus J(M ) corresponds to J(M/N ), that is, ψ(J(M )) = J(M/N ). Since ψ(J(M )) = J(M )/N , the result follows. Section 9.8 1. at ∈ (at+1 ), so there exists b ∈ R such that at = bat+1 . Since R is an integral domain we have 1 = ba. 2. Let a be a nonzero element of the Artinian integral domain R. The sequence (a) ⊇ (a2 ) ⊇ . . . stabilizes, so for some t we have (at ) = (at+1 ). By Problem 1, a has an inverse, proving that R is a ﬁeld. 3. If P is a prime ideal of R, then R/P is an Artinian integral domain by (7.5.7) and (2.4.5). By Problem 2, R/P is a ﬁeld, so by (2.4.3), P is maximal. 4. We have P ∩I ∈ S and P ∩I ⊆ I, so by minimality of I, P ∩I = I. Thus P ⊇ I = ∩nj=1 Ij , so by (7.6.2), P ⊇ Ij for some j. But P and Ij are maximal ideals, hence P = Ij . 5. If z ∈ F , with z = x + y, x ∈ M, y ∈ N , deﬁne h : F → M/JM ⊕ N/JN by h(z) = (x + JM ) + (y + JN ). Then h is an epimorphism with kernel JM + JN = JF , and the result follows from the ﬁrst isomorphism theorem. 6. By (9.8.5), M/JM is an n-dimensional vector space over the residue ﬁeld k. Since F , hence F/JF , is generated by n elements, F/JF has dimension at most n over k. Thus N/JN must have dimension zero, and the result follows. 7. Multiplication of an element of I on the right by a polynomial f (X, Y ) amounts to multiplication by the constant term of f . Thus I is ﬁnitely generated as a right R-module iﬀ it is ﬁnitely generated as an abelian group. This is a contradiction, because as an abelian group, n I = ⊕∞ n=0 ZX Y. 8. By the Hilbert basis theorem, Z[X] is a Noetherian ring and therefore a Noetherian left Z[X]-module. The isomorphic copy Z[X]Y is also a Noetherian left Z[X]-module, hence so is R, by (7.5.8). A left ideal of R that is ﬁnitely generated as a Z[X]-module is ﬁnitely generated as an R-module, so R is left-Noetherian. 9. The set {x1 , . . . , xn } spans V , and therefore contains a basis. If the containment is proper, then by (9.8.5) part (ii), {x1 , . . . , xn } cannot be a minimal generating set, a contradiction. 10. In vector-matrix notation we have y = Ax and therefore y = Ax, where aij = aij + J. By Problem 9, x and y are bases, so that det A = 0. But under the canonical map of R onto k, det A maps to det A, and therefore det A cannot belong to the kernel of the map, namely J. But by (8.5.9), J is the ideal of nonunits of R, so det A is a unit. Section 10.1 1. This is exactly the same as the proof for groups in (1.1.1). 2. Any ring homomorphism on Q is determined by its values on Z (write m = (m/n)n and apply the homomorphism g to get g(m/n) = g(m)/g(n). Thus if gi = hi, then g coincides with h on Z, so g = h, proving that i is epic.

page 20 of Solutions Chapters 6-10

3. Let AZB be a shorthand notation for the composition of the unique morphism from A to the zero object Z, followed by the unique morphism from Z to B. If Z is another zero object, then AZB = (AZ Z)B = A(Z ZB) = A(Z B) = AZ B, as claimed. 4. If is = it, then f is = f it = 0, so is(= it) = ih where h is unique. Thus s and t must coincide with h, hence i is monic. 5. A kernel of a monic f : A → B is 0, realized as the zero map from a zero object Z to A. For f 0 = 0, and if f g = 0, then f g = f 0; since f is monic, g = 0. But then g can be factored through 0ZA . Similarly, a cokernel of the epic f : A → B is the zero map from B to a zero object. 6. If j = ih and i = jh , then i = ihh , so by uniqueness in part (2) of the deﬁnition of kernel (applied when g = i), hh , and similarly h h, must be the identity. 7. Deﬁne h : K → A by h(x) = 1 for all x. Since K is nontrivial, h cannot be injective, so that g = h. But f g = f h, since both maps take everything in K to the identity of B. 8. The kernel of a ring homomorphism is an ideal, but not a subring (since it does not contains the multiplicative identity). 9. Let f : A → B be a noninjective ring homomorphism. Let C be the set of pairs (x, y) in the direct product A × A such that f (x) = f (y). Since f is not injective, there is an element (x, y) of C with x = y. Thus if D = {(x, x) : x ∈ A}, then D ⊂ C. If g is the projection of A × A on the ﬁrst coordinate, and h is the projection on the second coordinate, then f (g(x, y)) = f (x) and f (h(x, y)) = f (y), so f g = f h on the ring C. But g and h disagree on the nonempty set C \ D, so f is not monic. 10. Let g be the canonical map of N onto N/f (M ), and let h : N → N/f (M ) be identically zero. Since gf sends everything to 0, we have gf = hf with g = h. Thus f is not epic. Section 10.2 1. Suppose that y is the product of the xi . By deﬁnition of product, if fi : x → xi for all i, there is a unique f : x → y such that pi f = fi . Since pi : y → xi , we have y ≤ xi . Moreover, if x ≤ xi for all i, then x ≤ y. Therefore y is a greatest lower bound of the xi . 2. No. For example, consider the usual ordering on the integers. 3. By duality, a coproduct of the xi , if it exists, is a least upper bound. 4. If x has order r and y has order s, then rs(x + y) = s(rx) + r(sy) = 0. Thus the sum of two elements of ﬁnite order also has ﬁnite order, and the result follows. 5. The key point is that if f is a homomorphism of a torsion abelian groupS, then f (S) is a also torsion [since Ai , we have nf (x) = f (nx)]. Thus in diagram (1) with A = f (S) ⊆ T (A). Since Ai is the product in the category of abelian groups, it follows that T (A) satisﬁes the universal mapping property and is therefore the product in the category of torsion abelian groups. 6. Given homomorphisms fj : Gj → H, we must lift the fj to a homomorphism from the free product to H. This is done via f (a1 · · · an ) = f1 (a1 ) · · · fn (an ). If ij is the inclusion map from Gj to ∗i Gi , then f (ij (aj )) = f (aj ) = fj (aj ), as required. 7. We have pi f = fi , where fi : G → Ci . The fi can be chosen to be surjective (e.g., take G to be the direct product of the Ci ), and it follows that the pi are surjective. 8. Since f : C1 → C, we have f (a1 ) = na for some positive integer n. Thus a1 = f1 (a1 ) = p1 f (a1 ) = p1 (na) = np1 (a) = na1 ; 0 = f2 (a1 ) = p2 f (a1 ) = p2 (na) = np2 (a) = na2 . 9. By Problem 8, the order of C1 divides n − 1, and the order of C2 divides n. There are many choices of C1 and C2 for which this is impossible. For example, let C1 and C2 be nontrivial p-groups for a ﬁxed prime p. Section 10.3 1. If f : x → y, then F f : F x → F y. By deﬁnition of the category of preordered sets, this statement is equivalent to x ≤ y =⇒ F x ≤ F y. Thus functors are order-preserving maps.

page 21 of Solutions Chapters 6-10

2. F must take the morphism associated with xy to the composition of the morphism associated with F x and the morphism associated with F y. In other words, F (xy) = F (x)F (y), that is, F is a homomorphism. 3. If β ∈ X ∗ , then (gf )∗ (β) = βgf and f ∗ g ∗ (β) = f ∗ (βg) = βgf . 4. To verify the functorial property, note that (gf )∗∗ (v ∗∗ ) = v ∗∗ (gf )∗ = v ∗∗ f ∗ g ∗ (by Problem 3) and

g ∗∗ f ∗∗ v ∗∗ = g ∗∗ (v ∗∗ f ∗ ) = v ∗∗ f ∗ g ∗ .

Thus (gf )∗∗ = g ∗∗ f ∗∗ . If f is the identity, then so is f ∗ , and consequently so is f ∗∗ . 5. f ∗∗ tV (v) = f ∗∗ (v) = vf ∗ , and if β ∈ W ∗ , then (vf ∗ )(β) = v(f ∗ β) = (f ∗ β)(v) = βf (v). But tW f (v) = f (v) where f (v)(β) = βf (v). 6. Groups form a subcategory because every group is a monoid and every group homomorphism is, in particular, a monoid homomorphism. The subcategory is full because every monoid homomorphism from one group to another is also a group homomorphism. 7(a). If two group homomorphisms are the same as set mappings, they are identical as homomorphisms as well. Thus the forgetful functor is faithful. But not every map of sets is a homomorphism, so the forgetful functor is not full. (b) Since (f, g) is mapped to f for arbitrary g, the projection functor is full but not faithful (except in some degenerate cases). Section 10.4 1. If a homomorphism from Z2 to Q takes 1 to x, then 0 = 1 + 1 → x + x = 2x. But 0 must be mapped to 0, so x = 0. 2. A nonzero homomorphism can be constructed with 0 → 0, 1 → 12 . Then 1 + 1 → 12 + 12 = 1 = 0 in Q/Z. 3. Since a trivial group cannot be mapped onto a nontrivial group, there is no way that F g can be surjective. 4. Let f be a homomorphism from Q to Z. If r is any rational number and m is a positive integer, then r r r f (r) = f ( + · · · + ) = mf ( ) m m m so r f (r) f( ) = . m m But if f (r) = 0, we can choose m such that f (r)/m is not an integer, a contradiction. Therefore f = 0. 5. By Problem 4, Hom(Q, Z) = 0. But Hom(Z, Z) = 0, so as in Problem 3, Gf cannot be surjective. 6. We have Z2 ⊗ Z ∼ = Z2 and Z2 ⊗ Q = 0 [1 ⊗

2m m m m =1⊗ =2⊗ =0⊗ = 0.] n 2n 2n 2n

Thus the map Hf cannot be injective. 7. Since f∗ = F f is injective, f α = 0 implies α = 0, so f is monic and hence injective. Since g∗ f∗ = 0, we have gf α = 0 for all α ∈ Hom(M, A). Take M = A and α = 1A to conclude that gf = 0, so that im f ⊆ ker g. Finally, take M = ker g and α : M → B the inclusion map. Then g∗ α = gα = 0, so α ∈ ker g∗ = im f∗ . Thus α = f β for some β ∈ Hom(M, A). Thus ker g = M = im α ⊆ im f . 8. If (3) is exact for all possible R-modules N , then (1) is exact. This is dual to the result of Problem 7, and the proof amounts to interchanging injective and surjective, monic and epic, inclusion map and canonical map, kernel and cokernel.

page 22 of Solutions Chapters 6-10

Section 10.5

1. If x ∈ P , then f (x) can be expressed as i ti ei (a ﬁnite sum), and we deﬁne fi (x) = ti and xi = π(ei ). Then x = π(f (x)) = π( ti ei ) = ti π(ei ) = fi (x)xi . i

i

i

2. π(f (x)) = i fi (x)π(ei ) = i fi (x)xi = x. 3. By Problem 2, the exact sequence 0 → ker π → F → P → 0 (with π : F → P ) splits, and therefore P is a direct summand of the free module F and hence projective. 4. Since Rn is free, the “if” part follows from (10.5.3), part (4). If P is projective, then by the proof of (3) implies (4) in (10.5.3), P is a direct summand of a free module of rank n. [The free module can be taken to have a basis whose size is the same as that of a set of generators for P .] 5. This follows from Problem 1 with F = Rn . 6. If P is projective and isomorphic to M/N , we have an exact sequence 0 → N → M → P → 0. Since P is projective, the sequence splits by (10.5.3), part (3), so P is isomorphic to a direct summand of M . Conversely, assume that P is a direct summand of every module of which it is a quotient. Since P is a quotient of a free module F , it follows that P is a direct summand of F . By (10.5.3) part (4), P is projective. 7. In the diagram above (10.5.1), take M = R1 ⊕ R2 , with R1 = R2 = R. Take P = N = R1 , f = 1R1 , and let g be the natural projection of M on N . Then we can take h(r) = r + s, r ∈ R1 , s ∈ R2 , where s is either 0 or r. By replacing M by an arbitrary direct sum of copies of R, we can produce two choices for the component of h(r) in each Ri , i = 2, 3, . . . (with the restriction that only ﬁnitely many components are nonzero). Thus there will be inﬁnitely many possible choices for h altogether. Section 10.6 1. f g(a) = (g(a), 0) + W and g f (a) = (0, f (a)) + W , with (g(a), 0) − (0, f (a)) = (g(a), −f (a)) ∈ W . Thus f g = g f . 2. If (b, c) ∈ W , then b = g(a), c = −f (a) for some a ∈ A. Therefore g (c) + f (b) = −g f (a) + f g(a) = 0 and h is well-deﬁned. 3. hg (c) = h((0, c)+W ) = g (c)+0 = g (c) and hf (b) = h((b, 0)+W ) = 0+f (b) = f (b). 4. h ((b, c) + W ) = h ((0, c) + W + (b, 0) + W ) = h (g (c) + f (b)) = h g (c) + h f (b) = g (c) + f (b) = h((b, c) + W ). 5. If f (b) = 0, then by deﬁnition of f , (b, 0) ∈ W , so for some a ∈ A we have b = g(a) and f (a) = 0. Since f is injective, a = 0, hence b = 0 and f is injective. 6. If b ∈ B, c ∈ C, then surjectivity of f gives c = f (a) for some a ∈ A. Thus f (b + g(a)) = (b + g(a), 0) + W = (b + g(a), 0) + W + (−g(a), f (a)) + W [note that (−g(a), f (a)) ∈ W ] = (b, f (a)) + W = (b, c) + W , proving that f is surjective. 7. If (a, c) ∈ D, then f (a) = g(c) and f g (a, c) = f (a), gf (a, c) = g(c). Thus f g = gf . 8. If x ∈ E, then f g (x) = gf (x), so (g (x), f (x)) ∈ D. Take h(x) = (g (x), f (x)), which is the only possible choice that satisﬁes g h = g and f h = f . 9. If (a, c) ∈ D and f (a, c) = 0, then c = 0, so f (a) = g(c) = 0. Since f is injective, a = 0. Consequently, (a, c) = 0 and f is injective. 10. If c ∈ C, then there exists a ∈ A such that f (a) = g(c). Thus (a, c) ∈ D and f (a, c) = c, proving that f is surjective. 11. If x, y ∈ I, then f (xy) = xf (y) and f (xy) = f (yx) = yf (x). Thus xf (y) = yf (x), and if x and y are nonzero, the result follows upon division by xy. 12. We must extend f : I → Q to h : R → Q. Let z be the common value f (x)/x, x ∈ I, x = 0. Deﬁne h(r) = rz, r ∈ R. Then h is an R-homomorphism, and if x ∈ I, x = 0,

page 23 of Solutions Chapters 6-10

then h(x) = xz = xf (x)/x = f (x). Since h(0) = f (0) = 0, h is an extension of f and the result follows from (10.6.4). Section 10.7 1. The only nonroutine veriﬁcation is the check that sf ∈HomR (M, N ): (sf )(rm) = f (rms) = rf (ms) = r[(sf )(m)]. 2. (f r)(ms) = f (r(ms)) = f ((rm)s) = f (rm)s = [(f r)(m)]s. 3. (sf )(mr) = s(f (mr)) = s(f (m)r) = (sf (m)r = [(sf )(m)]r. 4. (f r)(sm) = (f (sm))r = (sf (m))r = s(f (m)r) = s[(f r)(m)]. 5. In Problem 2, M and N are right S-modules, so we write f on the left: (f r)m = f (rm). In Problem 3, M and N are right R-modules, and we write f on the left: (sf )m = s(f m). In Problem 4, M and N are left S-modules, and we write f on the right: m(f r) = (mf )r. 6. Let y ∈ M, r ∈ R with r = 0. By hypothesis, x = 1r y ∈ M , so we have x ∈ M such that y = rx, proving M divisible. 7. If y ∈ M, r ∈ R with r = 0, we must deﬁne 1r y. Since M is divisible, there exists x ∈ M such that y = rx, and we take 1r y = x. If x ∈ M and y = rx , then r(x − x ) = 0, and since M is torsion-free, x = x . Thus x is unique and scalar multiplication is well-deﬁned. 8. Let f be a nonzero R-homomorphism from Q to R. Then f (u) = 1 for some u ∈ Q. [If f (x) = r = 0, then rf (x/r) = f (rx/r) = f (x) = r, so we can take u = x/r.] Now if s is a nonzero element of R, then sf (u/s) = f (su/s) = f (u) = 1, so f (u/s) is an inverse of s. Consequently, R is a ﬁeld, contradicting the hypothesis. Section 10.8 1. Let A = R[X] where R is any commutative ring. As an R-algebra, A is generated by X, but A is not ﬁnitely generated as an R-module since it contains polynomials of arbitrarily high degree. 2. The bilinear map determined by (X i , Y j ) → X i Y j induces an R-homomorphism of R[X] ⊗R R[Y ] onto R[X, Y ], with inverse determined by X i Y j → X i ⊗ Y j . 3. Abbreviate X1 , . . . , Xn by X and Y1 , . . . , Ym by Y . Let A be a homomorphic image of R[X] under f , and B a homomorphic image of R[Y ] under g. Then A⊗R B is a homomorphic image of R[X] ⊗ R[Y ](∼ = R[X, Y ] by Problem 2) under f ⊗ g. 4. If f : A → B is an injective R-module homomorphism, then by hypothesis, (1 ⊗ f ) : S ⊗R A → S ⊗R B is injective. Also by hypothesis, (1 ⊗ (1 ⊗ f )) : M ⊗S S ⊗R A → M ⊗S S ⊗R B is injective. Since M ⊗S S ∼ = M , the result follows. 5. Let f : A → B be injective. Since A ⊗S S ∼ = A and B ⊗S S ∼ = B, it follows from the hypothesis that (f ⊗ 1) : A ⊗S (S ⊗R M ) → B ⊗S (S ⊗R M ) is injective. Thus S ⊗R M is a ﬂat S-module. 6. α is derived from the bilinear map S −1 R × M → S −1 M given by (r/s, x) → rx/s. We must also show that β is well-deﬁned. If x/s = y/t, then there exists u ∈ S such that utx = usy. Thus 1 ut 1 1 1 ⊗x= ⊗x= ⊗ utx = ⊗ usy = ⊗ y s sut sut sut t as required. By construction, α and β are inverses of each other and yield the desired isomorphism. 7. We must show that S −1 R ⊗R is an exact functor. But in view of Problem 6, an equivalent statement is the localization functor S −1 is exact, and this has already been proved in Section 8.5, Problem 5.

page 24 of Solutions Chapters 6-10

Section 10.9 1. The proof of (10.9.4) uses the fact that we are working in the category of modules. To simply say “duality” and reverse all the arrows, we would need an argument that did not depend on the particular category. 2. Let N be the direct limit of the Ni . The direct system {Ni , h(i, j)} induces a direct system {M ⊗ Ni , 1 ⊗ h(i, j)}. Compatibility in the new system reduces to compatibility in the old system; tensoring with 1 is harmless. Since compatible maps fi : Ni → B can be lifted to f : N → B, it follows that compatible maps gi : M ⊗ Ni → B can be lifted to g : M ⊗ N → B. Thus M ⊗ N satisﬁes the universal mapping property for {M ⊗ Ni }. 3. The direct limit is A = ∪∞ n=1 An , with αn : An → A the inclusion map 5. Each R-homomorphism f from the direct sum of the Ai to B induces an R-homomorphism fi : Ai → B. [fi is the injection of Ai into the direct sum, followed by f ]. Take α(f ) = (fi , i ∈ I) ∈ i HomR (Ai , B). Conversely, given such a family (fi , i ∈ I), the fi can be lifted uniquely to β(fi , i ∈ I) = f . Since α and β are inverse R-homomorphisms, the result follows. 6. If f : A → i Bi , deﬁne α(f ) = (pi f, i ∈ I) ∈ i HomR (A, Bi ), where pi is the projection of the direct product onto the ith factor. Conversely, given (gi , i ∈ I), where gi : A → Bi , the gi can be lifted to a unique g : A → i Bi such that pi g = gi for all i. If we take β(gi , i ∈ I) = g, then α and β are inverse R-homomorphisms, and the result follows. 7. There is a free module F such that F = M ⊕ M , and since F is torsion-free, so is M . Since M is injective, it is divisible, so by Problem 7 of Section 10.7, M is a vector space over the quotient ﬁeld Q, hence a direct sum of copies of Q. Therefore, using Problem 5 above and Problem 8 of Section 10.7, HomR (M, R) = HomR (⊕Q, R) ∼ =

Hom(Q, R) = 0.

8. By Problem 7, HomR (M, R) = 0. Let M be a direct summand of the free module F with basis {ri , i ∈ I}. If x is a nonzero element of M , then x has some nonzero coordinate with respect to the basis, say coordinate j. If pj is the projection of F on coordinate j (the j th copy of R), then pj restricted to M is a nonzero R-homomorphism from M to R. (Note that x does not belong to the kernel of pj .) Thus the assumption that M = 0 leads to a contradiction.

page 1 of Enrichment

Enrichment Chapters 1-4 form an idealized undergraduate course, written in the style of a graduate text. To help those seeing abstract algebra for the ﬁrst time, I have prepared this section, which contains advice, explanations and additional examples for each section in the ﬁrst four chapters. Section 1.1 When we say that the rational numbers form a group under addition, we mean that rational numbers can be added and subtracted, and the result will inevitably be rational. Similarly for the integers, the real numbers, and the complex numbers. But the integers (even the nonzero integers) do not form a group under multiplication. If a is an integer other than ±1, there is no integer b such that ab = 1. The nonzero rational numbers form a group under multiplication, as do the nonzero reals and the nonzero complex numbers. Not only can we add and subtract rationals, we can multiply and divide them (if the divisor is nonzero). The rational, reals and complex numbers are examples of ﬁelds, which will be studied systematically in Chapter 3. Here is what the generalized associative law is saying. To compute the product of the elements a, b, c, d and e, one way is to ﬁrst compute bc, then (bc)d, then a((bc)d), and ﬁnally [a((bc)d)e]. Another way is (ab), then (cd), then (ab)(cd), and ﬁnally ([(ab)(cd)]e). All procedures give the same result, which can therefore be written as abcde. Notice that the solution to Problem 6 indicates how to construct a formal proof of 1.1.4. Section 1.2 Groups whose descriptions diﬀer may turn out to be isomorphic, and we already have an example from the groups discussed in this section. Consider the dihedral group D6 , with elements I, R, R2 , F, RF, R2 F . Let S3 be the group of all permutations of {1, 2, 3}. We claim that D6 and S3 are isomorphic. This can be seen geometrically if we view D6 as a group of permutations of the vertices of an equilateral triangle. Since D6 has 6 elements and there are exactly 6 permutations of 3 symbols, we must conclude that D6 and S3 are essentially the same. To display an isomorphism explicitly, let R correspond to the permutation (1,2,3) and F to (2,3). Then I = (1), R = (1, 2, 3), R2 = (1, 3, 2), F = (2, 3), RF = (1, 2), R2 F = (1, 3). If G is a nonabelian group, then it must have an element of order 3 or more. (For example, S3 has two elements of order 3.) In other words, if every element of G has order 1 or 2, then G is abelian. To prove this, let a, b ∈ G; we will show that ab = ba. We can assume with loss of generality that a = 1 and b = 1. But then a2 = 1 and b2 = 1, so that a is its own inverse, and similarly for b. If ab has order 1, then ab = 1, so a and b are inverses of each other. By uniqueness of inverses, a = b, hence ab = ba. If ab has order 2, then abab = 1, so ab = b−1 a−1 = ba. Section 1.3 Here is another view of cosets that may be helpful. Suppose that a coded message is to be transmitted, and the message is to be represented by a code word x (an n-dimensional vector with components in some ﬁeld). The allowable code words are solutions of Ax = 0, where A is an m by n matrix, hence the set H of code words is an abelian group under componentwise addition, a subgroup of the abelian group G of all n-dimensional vectors. (In fact, G and H are vector space, but let’s ignore the additional structure.) Transmission is aﬀected by noise, so that the received vector is of the form z = x + y, where y is another n-dimensional vector, called the error vector or error pattern vector. Upon receiving z, we calculate the syndrome s = Az. If s turns out to be the zero vector, we declare that no

page 2 of Enrichment

error has occurred, and the transmitted word is z. Of course our decision may be incorrect, but under suitable assumptions about the nature of the noise, our decision procedure will minimize the probability of making a mistake. Again, let’s ignore this diﬃculty and focus on the algebraic aspects of the problem. We make the following claim: Two vectors z1 and z2 have the same syndrome if and only if they lie in the same coset of H in G. To prove this, observe that Az1 = Az2 iﬀ A(z1 − z2 ) = 0 iﬀ z1 − z2 ∈ H iﬀ z1 ∈ z2 + H. (We are working in an abelian group, so we use the additive notation z2 + H rather than the multiplicative notation z2 H.) Now suppose that we agree that we are going to correct the error pattern y1 , in other words, if we receive z = x1 + y1 , where x1 is a code word, we will decode z as x1 . If we receive z = x1 + y1 , where x1 is another code word, we decode z as x1 . Thus our procedure corrects y1 regardless of the particular word transmitted. Here is a key algebraic observation: If y1 and y2 are distinct vectors that lie in the same coset, it is impossible to correct both y1 and y2 . This holds because y1 = y2 + x for some code word x = 0, hence y1 + 0 = y2 + x. Therefore we cannot distinguish between the following two possibilities: 1. The zero word is transmitted and the error pattern is y1 ; 2. x is transmitted and the error pattern is y2 . It follows that among all vectors in a given coset, equivalently among all vectors having the same syndrome, we can choose exactly one as a correctable error pattern. If the underlying ﬁeld has only two elements 0 and 1, then (under suitable assumptions) it is best to choose to correct the pattern of minimum weight, that is, minimum number of 1’s. In particular, if the coset is the subgroup H itself, then we choose the zero vector. This agrees with our earlier proposal: if the received vector z has zero syndrome, we decode z as z itself, thus “correcting” the zero pattern, in other words, declaring that there has been no error in transmission. For further discussion and examples, see Information Theory by R.B. Ash, Dover 1991, Chapter 4. Section 1.4 Here are some intuitive ideas that may help in visualizing the various isomorphism theorems. In topology, we can turn the real interval [0, 1] into a circle by gluing the endpoints together, in other words identifying 0 and 1. Something similar is happening when we form the quotient group G/N where N is a normal subgroup of G. We have identiﬁed all the elements of N , and since the identity belongs to every subgroup, we can say that we have set everything in N equal to 1 (or 0 in the abelian case). Formally, (1.4.6) gives a correspondence between the subgroup of G/N consisting of the identity alone, and the subgroup N of G. We have already seen an example of this identiﬁcation process. In (1.2.4), we started with the free group G generated by the symbols R and F , and identiﬁed all sequences satisfying the relations Rn = I, F 2 = I, and RF = F R−1 (equivalently RF RF = I). Here we would like to take N to be the subgroup of G generated by Rn , F 2 , and RF RF , but N might not be normal. We will get around this technical diﬃculty when we discuss generators and relations in more detail in Section 5.8. Section 1.5 Direct products provide a good illustration of the use of the ﬁrst isomorphism theorem. Suppose that G = H × K; what can we say about G/H? If (h, k) ∈ G, then (h, k) =

page 3 of Enrichment

(h, 1K )(1H , k), and intuitively we have identiﬁed (h, 1K ) with the identity (1H , 1K ). What we have left is (1H , k), and it appears that G/H should be isomorphic to K. To prove this formally, deﬁne f : G → K by f (h, k) = k. Then f is an epimorphism whose kernel is {(h, 1K ) : h ∈ H}, which can be identiﬁed with H. By the ﬁrst isomorphism theorem, G/H ∼ = K. Section 2.1 Here is an interesting ring that will come up in Section 9.5 in connection with group representation theory. Let G = {x1 , . . . xm } be a ﬁnite group, and let R be any ring. The group ring RG consists of all elements r1 x1 + · · · + rm xm . Addition of elements is componentwise, just as if the xi were basis vectors of a vector space and the ri were scalars in a ﬁeld. Multiplication in RG is governed by the given multiplication in R, along with linearity. For example, (r1 x1 + r2 x2 )(s1 x1 + s2 x2 ) = r1 s1 x21 + r1 s2 x1 x2 + r2 s1 x2 x1 + r2 s2 x22 . The elements x21 , x1 x2 , x2 x1 , and x22 belong to G, which need not be abelian. The elements r1 s1 , r1 s2 , r2 s1 , and r2 s2 belong to R, which is not necessarily commutative. Thus it is essential to keep track of the order in which elements are written. Section 2.2 Here is some additional practice with ideals in matrix rings. If I is an ideal of Mn (R), we will show that I must have the form Mn (I0 ) for some unique ideal I0 of R. [Mn (I0 ) is the set of all n by n matrices with entries in I0 .] We note ﬁrst that for any matrix A, we have Eij AEkl = ajk Eil . This holds because Eij A puts row j of A in row i, and AEkl puts column k of A in column l. Thus Eij AEkl puts ajk in the il position, with zeros elsewhere. If I is an ideal of Mn (R), let I0 be the set of all entries a11 , where A = (aij ) is a matrix in I. To verify that I0 is an ideal, observe that (A + B)11 = a11 + b11 , ca11 = (cE11 A)11 , and a11 c = (AE11 c)11 . We will show that I = Mn (I0 ). If A ∈ I, set i = l = 1 in the basic identity involving the elementary matrices Eij (see the second paragraph above) to get ajk E11 ∈ I. Thus ajk ∈ I0 for all j and k, so A ∈ Mn (I0 ). Conversely, let A ∈ Mn (I0 ), so that ail ∈ I0 for all i, l. By deﬁnition of I0 , there is a matrix B ∈ I such that b11 = ail . Take j = k = 1 in the basic identity to get Ei1 BE1l = b11 Eil = ail Eil . Consequently, ail Eil ∈ I for all i and l. But the sum of the matrices ail Eil over all i and l is simply A, and we conclude that A ∈ I. To prove uniqueness, suppose that Mn (I0 ) = Mn (I1 ). If a ∈ I0 , then aE11 ∈ Mn (I0 ) = Mn (I1 ), so a ∈ I1 . A symmetrical argument completes the proof. Section 2.3 If a and b are relatively prime integers, then ai and bj are relatively prime for all positive integers i and j. Here is an analogous result for ideals. Suppose that the ideals I1 and I2 of the ring R are relatively prime, so that I1 + I2 = R. Let us prove that I12 and I2 are relatively prime as well. By the deﬁnitions of the sum and product of ideals, we have R = RR = (I1 + I2 )(I1 + I2 ) = I12 + I1 I2 + I2 I1 + I22 ⊆ I12 + I2 ⊆ R so R = I12 + I2 , as asserted. Similarly, we can show that R = I13 + I2 by considering the product of I12 + I2 and I1 + I2 . More generally, an induction argument shows that if I1 + . . . + In = R,

page 4 of Enrichment

then for all positive integers m1 , . . . , mn we have I1m1 + · · · + Inmn = R. Section 2.4 We have deﬁned prime ideals only when the ring R is commutative, and it is natural to ask why this restriction is imposed. Suppose that we drop the hypothesis of commutativity, and deﬁne prime ideals as in (2.4.4). We can then prove that if P is a prime ideal, I and J are arbitrary ideals, and P ⊇ IJ, then either P ⊇ I or P ⊇ J. [If the conclusion is false, there are elements a ∈ I \ P and b ∈ J \ P . Then ab ∈ IJ ⊆ P , but a ∈ / P and b ∈ / P, a contradiction.] If we try to reverse the process and show that a proper ideal P such that P ⊇ IJ implies P ⊇ I or P ⊇ J must be prime, we run into trouble. If ab belongs to P , then the principal ideal (ab) is contained in P . We would like to conclude that (a)(b) ⊆ P , so that (a) ⊆ P or (b) ⊆ P , in other words, a ∈ P or b ∈ P . But (ab) need not equal (a)(b). For example, to express the product of the element ar ∈ (a) and the element sb ∈ (b) as a multiple of ab, we must invoke commutativity. An explicit example: Let P be the zero ideal in the ring Mn (R) of n by n matrices over a division ring R (see Section 2.2, exercises). Since Mn (R) has no nontrivial two-sided ideals, P ⊇ IJ implies P ⊇ I or P ⊇ J. But ab ∈ P does not imply a ∈ P or b ∈ P , because the product of two nonzero matrices can be zero. This example illustrates another source of diﬃculty. The zero ideal P is maximal, but Mn (R)/P is not a division ring. Thus we cannot generalize (2.4.3) by dropping commutativity and replacing “ﬁeld” by “division ring”. [If R/M is a division ring, it does follow that M is a maximal ideal; the proof given in (2.4.3) works.] Section 2.5 Let’s have a brief look at polynomials in more than one variable; we will have much more to say in Chapter 8. For example, a polynomial f (X, Y, Z) in 3 variables is a sum of monomials; a monomial is of the form aX i Y j Z k where a belongs to the underlying ring R. The degree of such a monomial is i + j + k, and the degree of f is the maximum monomial degree. Formally, we can deﬁne R[X, Y ] as (R[X])[Y ], R[X, Y, Z] as (R[X, Y ])[Z], etc. Let f be a polynomial of degree n in F [X, Y ], where F is a ﬁeld. There are many cases in which f has inﬁnitely many roots in F . For example, consider f (X, Y ) = X + Y over the reals. The problem is that there is no direct extension of the division algorithm (2.5.1) to polynomials in several variables. The study of solutions to polynomial equations in more than one variable leads to algebraic geometry, which will be introduced in Chapter 8. Section 2.6 We have shown in (2.6.8) that every principal ideal domain is a unique factorization domain. Here is an example of a UFD that is not a PID. Let R = Z[X], which will be shown to be a UFD in (2.9.6). Let I be the maximal ideal < 2, X > (see (2.4.8)). If I is principal, then I consists of all multiples of a polynomial f (X) with integer coeﬃcients. Since 2 ∈ I, we must be able to multiply f (X) = a0 + a1 X + · · · + an X n by a polynomial g(X) = b0 + b1 X + · · · + bm X m and produce 2. There is no way to do this unless f (X) = 1 or 2. But if f (X) = 1 then I = R, a contradiction (a maximal ideal must be proper). Thus f (X) = 2, but we must also be able to multiply f (X) by some polynomial in Z[X] to produce X. This is impossible, and we conclude that I is not principal. A faster proof that Z[X] is not a PID is as follows. In (2.4.8) we showed that < 2 > and < X > are prime ideals that are not maximal, and the result follows from (2.6.9). On the other hand, the ﬁrst method produced an explicit example of an ideal that is not principal.

page 5 of Enrichment

Section 2.7 It may be useful to look at the Gaussian integers in more detail, and identify the primes in Z[i]. To avoid confusion, we will call a prime in Z[i] a Gaussian prime and a prime in Z a rational prime. Anticipating some terminology from algebraic number theory, we deﬁne the norm of the Gaussian integer a + bi as N (a + bi) = a2 + b2 . We will outline a sequence of results that determine exactly which Gaussian integers are prime. We use Greek letters for members of Z[i] and roman letters for ordinary integers. 1. α is a unit in Z[i] iﬀ N (α) = 1. Thus the only Gaussian units are ±1 and ±i. [If αβ = 1, then 1 = N (1) = N (α)N (β), so both N (α) and N (β) must be 1.] Let n be a positive integer. 2. If n is a Gaussian prime, then n is a rational prime not expressible as the sum of two squares. [n is a rational prime because any factorization in Z is also a factorization in Z[i]. If n = x2 + y 2 = (x + iy)(x − iy), then either x + iy or x − iy is a unit. By (1), n = 1, a contradiction.] Now assume that n is a rational prime but not a Gaussian prime. 3. If α = a + bi is a nontrivial factor of n, then gcd(a, b) = 1. [If the greatest common divisor d is greater than 1, then d = n. Thus α divides n and n divides α, so n and α are associates, a contradiction. ] 4. n is a sum of two squares. [Let n = (a + bi)(c + di); since n is real we have ad + bc = 0, so a divides bc. By (3), a and b are relatively prime, so a divides c, say c = ka. Then b(ka) = bc = −ad, so d = −bk. Thus n = ac − bd = ka2 + kb2 = k(a2 + b2 ). But a + bi is a nontrivial factor of n, so a2 + b2 = N (a + bi) > 1. Since n is a rational prime, we must have k = 1 and n = a2 + b2 .] By the above results, we have: 5. If n is a positive integer, then n is a Gaussian prime if and only if n is a rational prime not expressible as the sum of two squares. Now assume that α = a + bi is a Gaussian integer with both a and b nonzero. (The cases in which a or b is 0 are covered by (1) and (5).) 6. If N (α) is a rational prime, then α is a Gaussian prime. [If α = βγ where β and γ are not units, then N (α) = N (β)N (γ), where N (β) and N (γ) are greater than 1, contradicting the hypothesis.] Now assume that α is a Gaussian prime. 7. If N (α) = hk is a nontrivial factorization, so that h > 1 and k > 1, then α divides either h or k. If, say, α divides h, then so does its complex conjugate α. [We have N (α) = a2 + b2 = (a + bi)(a − bi) = αα = hk. Since α divides the product hk, it must divide one of the factors. If αβ = h, take complex conjugates to conclude that αβ = h.] 8. N (α) is a rational prime. [If not, then we can assume by (7) that α and α divide h. If α and α are not associates, then N (α) = αα divides h, so hk divides h and therefore k = 1, a contradiction. If α and its conjugate are associates, then one is ±i times the other. The only way this can happen is if α = γ(1 ± i) where γ is a unit. But then N (α) = N (γ)N (1 ± i) = N (1 ± i) = 2, a rational prime.]

page 6 of Enrichment

By the above results, we have: 9. If α = a + bi with a = 0, b = 0, then α is a Gaussian prime if and only if N (α) is a rational prime. The assertions (5) and (9) give a complete description of the Gaussian primes, except that it would be nice to know when a rational prime p can be expressed as the sum of two squares. We have 2 = 12 + 12 , so 2 is not a Gaussian prime, in fact 2 = (1 + i)(1 − i). If p is an odd prime, then p is a sum of two squares iﬀ p ≡ 1 mod 4, as we will prove at the beginning of Chapter 7. Thus we may restate (5) as follows: 10. If n is a positive integer, then n is a Gaussian prime iﬀ n is a rational prime congruent to 3 mod 4. [Note that a number congruent to 0 or 2 mod 4 must be even.] Section 2.8 Suppose that R is an integral domain with quotient ﬁeld F , and g is a ring homomorphism from R to an integral domain R . We can then regard g as mapping R into the quotient ﬁeld F of R . It is natural to try to extend g to a homomorphism g : F → F . If a, b ∈ R with b = 0, then a = b(a/b), so we must have g(a) = g(b)g(a/b). Thus if an extension exists, it must be given by g(a/b) = g(a)[g(b)]−1 . For this to make sense, we must have g(b) = 0 whenever b = 0, in other words, g is a monomorphism. [Note that if x, y ∈ R and g(x) = g(y), then g(x − y) = 0, hence x − y = 0, so x = y.] We will see in (3.1.2) that any homomorphism of ﬁelds is a monomorphism, so this condition is automatically satisﬁed. We can establish the existence of g by deﬁning it as above and then showing that it is a well-deﬁned ring homomorphism. This has already been done in Problem 8. We are in the general situation described in Problem 7, with S taken as the set of nonzero elements of R. We must check that g(s) is a unit in F for every s ∈ S, but this holds because g(s) is a nonzero element of F . Section 2.9 Here is another useful result relating factorization over an integral domain to factorization over the quotient ﬁeld. Suppose that f is a monic polynomial with integer coeﬃcients, and that f can be factored as gh, where g and h are monic polynomials with rational coeﬃcients. Then g and h must have integer coeﬃcients. More generally, let D be a unique factorization domain with quotient ﬁeld F , and let f be a monic polynomial in D[X]. If f = gh, with g, h ∈ F [X], then g and h must belong to D[X]. To prove this, we invoke the basic proposition (2.9.2) to produce a nonzero λ ∈ F such that λg ∈ D[X] and λ−1 h ∈ D[X]. But g and h are monic, so λ = 1 and the result follows. Let f be a cubic polynomial in F [X]. If f is reducible, it must have a linear factor and hence a root in F . We can check this easily if F is a ﬁnite ﬁeld; just try all possibilities. A ﬁnite check also suﬃces when F = Q, by the rational root test (Section 2.9, Problem 1). If g is a linear factor of f , then f /g = h is quadratic. We can factor h as above, and in addition the quadratic formula is available if square roots can be extracted in F . In other words, if a ∈ F , then b2 = a for some b ∈ F . Section 3.1 All results in this section are basic and should be studied carefully. You probably have some experience with polynomials over the rational numbers, so let’s do an example with a rather diﬀerent ﬂavor. Let F = F2 be the ﬁeld with two elements 0 and 1, and let f ∈ F [X] be the polynomial X 2 + X + 1. Note that f is irreducible over F , because if f

page 7 of Enrichment

were factorable, it would have a linear factor and hence a root in F . This is impossible, as f (0) = f (1) = 1 = 0. If we adjoin a root α of f to produce an extension F (α), we know that f is the minimal polynomial of α over F , and that F (α) consists of all elements b0 + b1 α, with b0 and b1 in F . Since b0 and b1 take on values 0 and 1, we have constructed a ﬁeld F (α) with 4 elements. Moreover, all nonzero elements of F (α) can be expressed as powers of α, as follows: α0 = 1, α1 = α, α2 = −α − 1 = 1 + α. (The last equality follows because 1+1=0 in F .) This is a typical computation involving ﬁnite ﬁelds, which will be studied in detail in Chapter 6. Section 3.2 We found in Problem 3 that a splitting ﬁeld for X 4 − 2 has degree 8 over Q. If we make a seemingly small change and consider f (X) = X 4 − 1, the results are quite diﬀerent. The roots of f are 1, i, −1 and −i. Thus Q(i) is the desired splitting ﬁeld, and it has degree 2 over Q because the minimal polynomial of i over Q has degree 2. A general problem suggested by this example is to describe a splitting ﬁeld for X n − 1 over Q for an arbitrary positive integer n. The splitting ﬁeld is Q(ω), where ω is a primitive nth root of unity, for example, ω = ei2π/n . We will see in Section 6.5 that the degree of Q(ω) over Q is ϕ(n), where ϕ is the Euler phi function. Section 3.3 In Problem 8 we used the existence of an algebraic closure of F to show that any set of nonconstant polynomials in F [X] has a splitting ﬁeld over F . Conversely, if we suppose that it is possible to ﬁnd a splitting ﬁeld K for an arbitrary family of polynomials over the ﬁeld F , then the existence of an algebraic closure of F can be established quickly. Thus let K be a splitting ﬁeld for the collection of all polynomials in f [X], and let C be the algebraic closure of F in K (see (3.3.4)). Then by deﬁnition, C is an algebraic extension of F and every nonconstant polynomial in F [X] splits over C. By (3.3.6), C is an algebraic closure of F . Section 3.4 Let’s have another look at Example 3.4.8 with p =√2 to get some additional practice with separability and inseparability. We have seen that t is not separable over F , in√fact it is purely inseparable because its minimal polynomial X 2√ − t can be written as (X − t)2 . 3 But if we adjoin a cube root of t, the resulting element t is separable over F , because X 3 − t has nonzero derivative, equivalently does not belong to F [X 2 ] (see 3.4.3). √ √ √ 6 also √ that adjoining t and 3 t is equivalent to√ adjoining t, in other √ √ √ words, √Notice 3 6 6 To see this, ﬁrst observe that if α = t, then t = α3 and 3 t = α2 . F ( t, t) = F ( t). √ √ On the other hand, ( t/ 3 t)6 = t. It is possible for an element α to be both separable and purely inseparable over F , but it happens if and only if α belongs to F . The minimal polynomial of α over F must have only one distinct root and no repeated roots, so min(α, F ) = X − α. But the minimal polynomial has coeﬃcients in F (by deﬁnition), and the result follows. Section 3.5 Suppose we wish to ﬁnd the Galois group of the extension E/F , where E = F (α). Assume that α is algebraic over F with minimal polynomial f , and that f has n distinct roots α1 = α, α2 , . . . , αn in some splitting ﬁeld. If σ ∈ Gal(E/F ), then σ permutes the roots of f by (3.5.1). Given any two roots αi and αj , i = j, we can ﬁnd an F -isomorphism that carries αi into αj ; see (3.2.3). Do not jump to the conclusion that all permutations are allowable, and therefore Gal(E/F ) is isomorphic to Sn . For example, we may not be

page 8 of Enrichment

able to simultaneously carry α1 into α2 and α3 into α4 . Another diﬃculty is that the F isomorphism carrying αi into αj need not be an F -automorphism of E. This suggests that normality of the extension is a key property. If E/F is the non-normal extension of Example (3.5.10), the only allowable permutation is the identity. Section 4.1 Finitely generated algebras over a commutative ring R frequently appear in applications to algebraic number theory and algebraic geometry. We say that A is a ﬁnitely generated R-algebra if there are ﬁnitely many elements x1 , . . . , xn in A such that every element of A is a polynomial f (x1 , . . . , xn ) with coeﬃcients in R. Equivalently, A is a homomorphic image of the polynomial ring R[X1 , . . . , Xn ]. The homomorphism is determined explicitly by mapping Xi to xi , i = 1, . . . , n. The polynomial f (X1 , . . . , Xn ) is then mapped to f (x1 , . . . , xn ). If every element is not just a polynomial in the xi but a linear combination of the xi with coeﬃcients in R, then A is a ﬁnitely generated module over R. To see the diﬀerence clearly, look at the polynomial ring R[X], which is a ﬁnitely generated R algebra. (In the above discussion we can take n = 1 and x1 = X.) But if f1 , . . . , fn are polynomials in R[X] and the maximum degree of the fi is m, there is no way to take linear combinations of the fi and produce a polynomial of degree greater than m. Thus R[X] is a not a ﬁnitely generated R-module. Section 4.2 Here is some practice working with quotient modules. Let N be a submodule of the R-module M , and let π be the canonical map from M onto M/N , taking x ∈ M to x + N ∈ M/N . Suppose that N1 and N2 are submodules of M satisfying (a) N1 ≤ N2 ; (b) N1 ∩ N = N2 ∩ N ; (c) π(N1 ) = π(N2 ). Then N1 = N2 . To prove this, let x ∈ N2 . Hypothesis (c) says that (N1 + N )/N = (N2 + N )/N ; we don’t write Ni /N, i = 1, 2, because N is not necessarily a submodule of N1 or N2 . Thus x + N ∈ (N2 + N )/N = (N1 + N )/N , so x + N = y + N for some y ∈ N1 . By (a), y ∈ N2 , hence x − y ∈ N2 ∩ N = N1 ∩ N by (b). Therefore x − y and y both belong to N1 , and consequently so does x. We have shown that N2 ≤ N1 , and in view of hypothesis (a), we are ﬁnished. Section 4.3 If M is a free R-module with basis S = (xi ), then an arbitrary function f from S to an arbitrary R-module N has a unique extension to an R-homomorphism f : M → N ; see (4.3.6). This property characterizes free modules, in other words, if M is an R-module homomorphism with a subset S satisfying the above property, then M is free with basis S. To see this, build a free module M with basis S = (yi ) having the same cardinality as S. For example, we can take M to be the direct sum of copies of R, as many copies as there are elements of S. Deﬁne f : S → S ⊆ M by f (xi ) = yi , and let f be the unique extension of f to an R-homomorphism from M to M . Similarly, deﬁne g : S → S ⊆ M by g(yi ) = xi , and let g be the unique extension of g to an R-homomorphism from M to M . Note that g exists and is unique because M is free. Now g ◦ f is the identity on S, so by uniqueness of extensions from S to M , g ◦ f is the identity on M . Similarly, f ◦ g is the identity on M . Thus M and M are not only isomorphic, but the isomorphism we have constructed carries S into S . It follows that M is free with basis S.

page 9 of Enrichment

This is an illustration of the characterization of an algebraic object by a universal mapping property. We will see other examples in Chapter 10. Section 4.4 Here is some practice in decoding abstract presentations. An R-module can be deﬁned as a representation of R in an endomorphism ring of an abelian group M . What does this mean? First of all, for each r ∈ R, we have an endomorphism fr of the abelian group M , given by fr (x) = rx, x ∈ M . To say that fr is an endomorphism is to say that r(x + y) = rx + ry, x, y ∈ M, r ∈ R. Second, the mapping r → fr is a ring homomorphism from R to EndR (M ). (Such a mapping is called a representation of R in EndR (M ).) This says that fr+s (x) = fr (x)+fs (x), frs (x) = fr (fs (x)), and f1 (x) = x. In other words, (r + s)x = rx + sx, (rs)x = r(sx), and 1x = x. Thus we have found a fancy way to write the module axioms. If you are already comfortable with the informal view of a module as a “vector space over a ring”, you are less likely to be thrown oﬀ stride by the abstraction. Section 4.5 The technique given in Problems 1-3 for ﬁnding new bases and generators is worth emphasizing. We start with a matrix A to be reduced to Smith normal form. The equations U = AX give the generators U of the submodule K in terms of the basis X of the free module M . The steps in the Smith calculation are of two types: 1. Premultiplication by an elementary row matrix R. This corresponds to changing generators via V = RU . 2. Postmultiplication by an elementary column matrix C. This corresponds to changing bases via Y = C −1 X. Suppose that the elementary row matrices appearing in the calculation are R1 , . . . , Rs , in that order, and the elementary column matrices are C1 , . . . , Ct , in that order. Then the matrices Q and P are given by Q = Rs · · · R 2 R 1 ,

P −1 = C1 C2 · · · Ct

hence P = Ct−1 · · · C2−1 C1−1 . The ﬁnal basis for M is Y = P X, and the ﬁnal generating set for K is V = QU = SY , where S = QAP −1 is the Smith normal form (see 4.4.2). Section 4.6 Here is a result that is used in algebraic number theory. Let G be a free abelian group of rank n, and H a subgroup of M . By the simultaneous basis theorem, there is a basis y1 , . . . yn of G and there are positive integers a1 , . . . ar , r ≤ n, such that ai divides ai+1 for all i, and a1 y1 , . . . , ar yr is a basis for H. We claim that the abelian group G/H is ﬁnite if and only if r = n, and in this case, the size of G/H is |G/H| = a1 a2 · · · ar . To see this, look at the proof of (4.6.3) with Rn replaced by G and K by H. The argument shows that G/H is the direct sum of cyclic groups Z/Zai , i = 1, . . . , n, with ai = 0 for r < i ≤ n. In other words, G/H is the direct sum of r ﬁnite cyclic groups (of order a1 , . . . , ar respectively) and n − r copies of Z. The result follows. Now assume that r = n, and let x1 , . . . , xn and z1 , . . . , zn be arbitrary bases for G and H respectively. Then each zi is a linear combination of the xi with integer coeﬃcients; in matrix form, z = Ax. We claim that |G/H| is the absolute value of the determinant of A. To verify this, ﬁrst look at the special case xi = yi and zi = ai yi , i = 1, . . . , n. Then A

page 10 of Enrichment

is a diagonal matrix with entries ai , and the result follows. But the special case implies the general result, because any matrix corresponding to a change of basis of G or H is unimodular, in other words, has determinant ±1. (See Section 4.4, Problem 1.) Section 4.7 Here is some extra practice in diagram chasing. The diagram below is commutative with exact rows. f g A → B → C → 0 t↓ A

u↓ → f

B

↓v → g

C

→

0

If t and u are isomorphisms, we will show that v is also an isomorphism. Let c ∈ C ; then c = g b for some b ∈ B . Since u is surjective, g b = g ub for some b ∈ B. By commutativity, g ub = vgb, which proves that v is surjective. Now assume vc = 0. Since g is surjective, c = gb for some b ∈ B. By commutativity, vgb = g ub = 0. Thus ub ∈ ker g = im f , so ub = f a for some a ∈ A . Since t is surjective, f a = f ta for some a ∈ A. By commutativity, f ta = uf a. We now have ub = uf a, so b − f a ∈ ker u, hence b = f a because u is injective. Consequently, c = gb = gf a = 0 which proves that v is injective.

page 1 of Endmatter

BIBLIOGRAPHY General Cohn, P.M., Algebra, Volumes 1 and 2, John Wiley and Sons, New York, 1989 Dummit, D.S. and Foote, R.M., Abstract Algebra, Prentice-Hall, Upper Saddle River, NJ, 1999 Hungerford, T.M., Algebra, Springer-Verlag, New York, 1974 Isaacs, I.M., Algebra, a Graduate Course, Brooks-Cole, a division of Wadsworth, Inc., Paciﬁc Grove, CA, 1994 Jacobson, N., Basic Algebra I and II, W.H. Freeman and Company, San Francisco, 1980 Lang, S., Algebra, Addison-Wesley, Reading, MA, 1993 Modules Adkins, W.A., and Weintraub, S.H., Algebra, An Approach via Module Theory, SpringerVerlag, New York, 1992 Blyth, T.S., Module Theory, Oxford University Press, Oxford, 1990 Basic Group Theory Alperin, J.L., and Bell, R.B., Groups and Representations, Springer-Verlag, New York, 1995 Humphreys, J.F., A Course in Group Theory, Oxford University Press, Oxford 1996 Robinson, D.S., A Course in the Theory of Groups, Springer-Verlag, New York, 1993 Rose, J.S., A Course on Group Theory, Dover, New York, 1994 Rotman, J.J., An Introduction to the Theory of Groups, Springer-Verlag, New York, 1998 Fields and Galois Theory Adamson, I.T., Introduction to Field Theory, Cambridge University Press, Cambridge, 1982 Garling, D.J.H., A Course in Galois Theory, Cambridge University Press, Cambridge, 1986 Morandi, P., Fields and Galois Theory, Springer-Verlag, New York, 1996 Roman, S., Field Theory, Springer-Verlag, New York, 1995 Rotman, J.J., Galois Theory, Springer-Verlag, New York, 1998 Algebraic Number Theory Borevich, Z.I., and Shafarevich, I.R., Number Theory, Academic Press, San Diego, 1966 Fr¨ ohlich, A., and Taylor, M.J., Algebraic Number Theory, Cambridge University Press, Cambridge, 1991 Gouvea, F.Q., p-adic Numbers, Springer-Verlag, New York, 1997 Janusz, G.J., Algebraic Number Fields, American Mathematical Society, Providence, 1996 Lang, S., Algebraic Number Theory, Springer-Verlag, New York, 1994 Marcus, D.A., Number Fields, Springer-Verlag, New York, 1977 Samuel, P., Algebraic Theory of Numbers, Hermann, Paris, 1970 Algebraic Geometry and Commutative Algebra Atiyah, M.F., and Macdonald, I.G., Introduction to Commutative Algebra, Addison-Wesley, Reading, MA, 1969 Bump, D., Algebraic Geometry, World Scientiﬁc, Singapore, 1998

page 2 of Endmatter

Cox, D., Little, J., and O’Shea, D., Ideals, Varieties, and Algorithms, Springer-Verlag, New York, 1992 Eisenbud, D., Commutative Algebra with a View Toward Algebraic Geometry, SpringerVerlag, New York, 1995 Fulton, W., Algebraic Curves, W.A. Benjamin, New York, 1969 Hartshorne, R., Algebraic Geometry, Springer-Verlag, New York, 1977 Kunz, E., Introduction to Commutative Algebra and Algebraic Geometry, Birkh¨ auser, Boston, 1985 Matsumura, H., Commutative Ring Theory, Cambridge University Press, Cambridge, 1986 Reid, M., Undergraduate Algebraic Geometry, Cambridge University Press, Cambridge, 1988 Reid, M., Undergraduate Commutative Algebra, Cambridge University Press, Cambridge, 1995 Shafarevich, I.R., Basic Algebraic Geometry, Volumes 1 and 2, Springer-Verlag, New York 1988 Ueno, K., Algebraic Geometry 1, American Mathematical Society, Providence 1999 Noncommutative Rings Anderson, F.W., and Fuller, K.R., Rings and Categories of Modules, Springer-Verlag, New York, 1992 Beachy, J.A., Introductory Lectures on Rings and Modules, Cambridge University Press, Cambridge, 1999 Farb, B., and Dennis, R.K., Noncommutative Algebra, Springer-Verlag, New York, 1993 Herstein, I.N., Noncommutative Rings, Mathematical Association of America, Washington, D.C., 1968 Lam, T.Y., A First Course in Noncommutative Rings, Springer-Verlag, New York, 1991 Group Representation Theory Curtis, C.W., and Reiner, I., Methods of Representation Theory, John Wiley and Sons, New York, 1981 Curtis, C.M., and Reiner, I., Representation Theory of Finite Groups and Associative Algebras, John Wiley and Sons, New York, 1966 Dornhoﬀ, L., Group Representation Theory, Marcel Dekker, New YOrk, 1971 James, G., and Liebeck, M., Representations and Characters of Groups, Cambridge University Press, Cambridge, 1993 Homological Algebra Hilton, P.J., and Stammbach, U., A Course in Homological Algebra, Springer-Verlag, New York, 1970 Hilton, P., and Wu, Y-C., A Course in Modern Algebra, John Wiley and Sons, New York, 1974 Mac Lane, S., Categories for the Working Mathematician, Springer-Verlag, New York, 1971 Rotman, J.J., An Introduction to Algebraic Topology, Springer-Verlag, New York, 1988 Rotman, J.J., An Introduction to Homological Algebra, Springer-Verlag, New York, 1979 Weibel, C.A., An Introduction to Homological Algebra, Cambridge University Press, Cambridge, 1994

page 3 of Endmatter

List of Symbols Throughout the text, ⊆ means subset, ⊂ means proper subset Zn integers modulo n 1.1 Z integers 1.1 < A > subgroup generated by A 1.1 Sn symmetric group 1.2 An alternating group 1.2 D2n dihedral group 1.2 ϕ Euler phi function 1.1, 1.3 normal subgroup 1.3 proper normal subgrouad 1.3 ker kernel 1.3, 2.2 ∼ = isomorphism 1.4 Z(G) center of a group 1.4 H × K direct product 1.5 Q rationals 2.1 Mn (R) matrix ring 2.1 R[X] polynomial ring 2.1 R[[X]] formal power series ring 2. End endomorphism ring 2.1 < X > ideal generated by X 2.2 UFD unique factorization domain 2.6 PID principal ideal domain 2.6 ED Euclidean domain 2.7 min(α, F ) minimal polynomial 3.1 K composite of ﬁelds 3.1 i i Gal(E/F ) Galois group 3.5 0 the module {0} and the ideal {0} 4.1 ⊕ i Mi direct sum of modules 4.3 sum of modules 4.3 i Mi HomR (M, N ) set of R-module homomorphisms from M to N 4.4 EndR (M ) endomorphism ring 4.4 g • x group action 5.1 G commutator subgroup 5.7 G(i) derived subgroups 5.7 < S | K > presentation of a group 5.8 F(H) ﬁxed ﬁeld 6.1 G(K) ﬁxing group 6.1 GF (pn ) ﬁnite ﬁeld with pn elements 6.4 Ψn (X) nth cyclotomic polynomial 6.5 ∆ product of diﬀerences of roots 6.6 D discriminant 6.6,7.4 N [E/F ] norm 7.3 T [E/F ] trace 7.3 char characteristic polynomial 7.3 nP (I) exponent of P in the factorization of I 7.7 vp p-adic valuation 7.9 | |p p-adic absolute value 7.9 V (S) variety in aﬃne space 8.1 I(X) ideal of a set of points 8.1 k[X1 , . . . , Xn ] polynomial ring in n variables over the ﬁeld k 8.1 √ I radical of an ideal 8.3 k(X1 , . . . , Xn ) rational function ﬁeld over k 8.4

page 4 of Endmatter

S −1 R localization of the ring R by S 8.5 S −1 M localization of the module M by S 8.5 N (R) nilradical of the ring R 8.6 M ⊗R N tensor product of modules 8.7 UMP universal mapping property 8.7 A ⊗ B tensor (Kronecker) product of matrices 8.7 kG group algebra 9.5 RG group ring 9.5 J(M ), J(R) Jacobson radical 9.7 HomR (M, ), HomR ( , N ) hom functors 10.3 M ⊗R , ⊗R N tensor fuctors 10.3 Q/Z additive group of rationals mod 1 10.6 (also 1.1, Problem 7) Z(p∞ ) quasicyclic group A10 G[n] elements of G annihilated by n A10 Hn homology functor S1 f g chain homotopy S1 ∂ connecting homomorphism S2, S3 P∗ → M projective resolution S4 M → E∗ injective resolution S4 Ln F left derived functor S5 Rn F right derived functor S5 Tor derived functor of ⊗ S5 Ext derived functor of Hom S5

page 5 of Endmatter

INDEX abelian category 10.4 abelian group 1.1 absolute value 7.9 action of a group on a set 5.1 adjoint associativity 10.7 adjoint functors 10.7 aﬃne n-space 8.1 aﬃne variety 8.1 AKLB setup 7.3 algebra 4.1 algebraic closure 3.3, 10.9 algebraic curve 8.3 algebraic element 3.1 algebraic extension 3.1 algebraic function ﬁeld 6.9 algebraic geometry 8.1ﬀ algebraic integers 7.1 algebraic number 3.3, 7.1 algebraic number theory 7.1ﬀ, 7.3 algebraically closed ﬁeld 3.3 algebraically independent set 6.9 algebraically spanning set 6.9 alternating group 1.2 annihilator 4.2, 9.2, 9.7 archimedian absolute value 7.9 Artin-Schreier theorem 6.7 Artinian modules 7.5 Artinian rings 7.5 ascending chain condition (acc) 2.6, 7.5 associates 2.6 associative law 1.1, 2.1 automorphism 1.3 Baer’s criterion 10.6 base change 10.8 basis 4.3 bilinear mapping 8.7 binomial expansion modulo p 3.4 binomial theorem 2.1 boundary S1 canonical map 1.3 category 10.1 Cauchy’s theorem 5.4 Cayley’s theorem 5.1 center of a group 1.4 center of a ring 4.1 central series 5.7 centralizer 5.2 chain complex S1 chain homotopy S1 chain map S1 chain rule 1.3, 3.1

page 6 of Endmatter

character 6.1 characteristic of a ring or ﬁeld 2.1 characteristic polynomial 7.3 characteristic subgroup 5.7 chief series 5.6 Chinese remainder theorem 2.3 class equation 5.2 cokernel 10.1 colorings 5.3 commutative diagram 1.4 commutative ring 2.1 commutator 5.7 compatible morphisms 10.9 complete ring of fractions 2.8 composite of ﬁelds 3.1, 6.2 composition factors 5.6, 7.5 composition length 5.6, 7.5 composition of morphisms 10.1 composition series 5.6, 7.5 conjugate elements 5.1, 5.2 conjugate subﬁelds 6.2 conjugate subgroups 5.1, 6.2 conjugates of a ﬁeld ement 3.5 conjugation 5.1, 5.2-1 connecting homomorphism S2, S3 constructible numbers and points 6.8 content 2.9 contravariant functor 10.3 coproduct 10.2 core 5.1 correspondence theorem for groups 1.4 correspondence theorem for modules 4.2 correspondence theorem for rings 2.3 coset 1.3 counting two ways 5.3 covariant functor 10.3 cycle 1.2, S1 cyclic extension 6.7 cyclic group 1.1 cyclic module 4.2, 9.1, 9.2, 9.7 cyclotomic extension.5 cyclotomic ﬁeld 6.5, 7.2 cyclotomic polynomial 6.5 decomposable module 9.6 Dedekind domain 7.6, 7.7 Dedekind’s lemma 6.1, 6.7, 7.3, 7.4 degree 2.5 deleted projective (or injective) resolution S4 derivative of a polynomial 3.4 derived functors S5 derived length 5.7 derived series 5.7 descending chain condition 7.5

page 7 of Endmatter

diagram chasing 4.7 diﬀerential S1 dihedral group 1.2, 5.8, 5.8 (inﬁnite dilhedral group), 9.5 direct limit 10.9 direct product of groups 1.5 direct product of modules 4.3 direct product of rings 2.3 direct sum of modules 4.3 direct system 10.9 directed set 10.9 discriminant 6.6, A6, 7.4 divides means contains 2.6, 7.7 divisible abelian group A10 divisible module 10.6 division ring 2.1, 9.1 double centralizer 9.2 double dual functor 10.3 dual basis 7.4 duality 10.1 duplicating the cube 6.8 Eisenstein’s irreducibility criterion 2.9 elementary divisors 4.6 elementary symmetric functions 6.1 embedding 3.3,3.5 embedding in an injective module 10.7 endomorphism 1.3,4.4 epic 10.1 epimorphism 1.3 equivalent absolute values 7.9 equivalent matrices 4.4 equivalent matrix representations 9.5 Euclidean domain 2.7 Euler’s identity 2.1 Euler’s theorem 1.3 evaluation map 2.1 exact functor 8.5, 10.4 exact sequence 4.7 exponent of a group 1.1, 6.4 Ext S5 extension of a ﬁeld 3.1 extension of scalars 8.7, 10.8 exterior algebra 8.8 F-isomorphism, etc. 3.2 factor theorem for groups 1.4 factor theorem for modules 4.2 factor theorem for rings 2.3 faithful action 5.1 faithful module 7.1, 9.2, 9.4 faithful representation 9.5 Fermat primes 6.8 Fermat’s little theorem 1.3 ﬁeld 2.1 ﬁeld discriminant 7.4

page 8 of Endmatter

ﬁnite abelian groups 4.6 ﬁnite extension 3.1 ﬁnite ﬁelds 6.4 ﬁnitely cogenerated module 7.5 ﬁnitely generated algebra 10.8 ﬁnitely generated module 4.4 ﬁnitely generated submodule 7.5 ﬁve lemma 4.7 ﬁxed ﬁeld 6.1 ﬁxing group 6.1 ﬂat modules 10.8 forgetful functor 10.3 formal power series 2.1, 8.2 four group 1.2, 1.5, A6 four lemma 4.7 fractional ideal 7.6 Frattini argument 5.8-2 free abelian gup functor 10.3 free group 5.8-1 free module 4.3,15 free product 10.2-2 Frobenius automorphism 3.4, 6.4 full functor 10.3 full ring of fractions 2.8 full subcategory 10.3 functor 10.3 fundamental decomposition theorem (for ﬁnitely generated modules over a PID) 4.6 fundamental theorem of Galois theory 6.2-1 Galois extension 3.5, 6.1ﬀ. Galois group 3.5, 6.1ﬀ Galois group of a cubic, 6.6 Galois group of a polynomial 6.3 Galois group of a quadratic 6.3 Galois group of a quartic A6 Gauss’ lemma 2.9 Gaussian integers 2.1, 2.7 general equation of degree n 6.8 general linear group 1.3 generating set 4.3 generators and relations 1.2, 4.6, 5.8 greatest common divisor 2.6, 7.7 group 1.1 group algebra 9.5 group representations 9.5 group ring 9.5 Hermite normal form 4.5 Hilbert basis theorem 8.2 Hilbert’s Nullstellensatz 8.3, 8.4 Hilbert’s Theorem 90 7.3 hom functors 10.3-1 homology functors S1 homology group S1 homology module S1

page 9 of Endmatter

homomorphism from R to M determined by what it does to the identity, 9.4, S6 homomorphism of algebras 4.1 homomorphism of groups 1.3 homomorphism of modules 4.1 homomorphism of rings 2.2 Hopkins-Levitzki theorem 9.8 hypersurface 8.2 ideal 2.2, 8.1 ideal class group 7.8 idempotent linear transformation 9.5 image 2.3, 4.1 indecomposable module 9.6 index 1.3 inductive limit 10.9 initial object 10.1 injection (inclusion) 4.7 injective hull 10.7 injective modules 10.6 injective resolution S4 inner automorphism 1.4, 5.7 integral basis 7.2, 7.4 integral closure 7.1 integral domain 2.1 integral extensions 7.1 integral ideal 7.6 integrally closed 7.1 invariant factors 4.5 inverse limit 10.9 inverse system 10.9 inversions 1.2 irreducible element 2.6 irreducible ideal 8.6 irreducible polynomial 2.9 irreducible variety 8.1 isomorphic groups 1.1 isomorphism 1.3 isomorphism extension theorem 3.2 isomorphism theorems for groups 1.4 isomorphism theorems for modules 4.2 isomorphism theorems for rings 2.3 Jacobson radical 9.7 Jacobson’s theorem 9.2 Jordan-Holder theorem 5.6, 7.5 kernel 1.3, 2.2, 10.1 kernel of an action 5.1 Kronecker product of matrices 8.7 Krull-Schmidt theorem 9.6 Kummer extension 6.7 Lagrange interpolation formula 2.5 Lagrange’s theorem 1.3 Laurent series 7.9 leading coeﬃcient 2.5

page 10 of Endmatter

least common multiple 2.6, 7.7 left adjoint 10.7 left cancellable 10.1 left derived functors S5 left exact functor 10.4 left ideal 2.2 left resolution S4 left-Noetherian ring 9.8 left-quasiregular element 9.7 left-semisimple ring 9.6 length of a module 7.5 lifting of a map 4.3, 10.2 linearly indepdent set 4.3 local ring 2.4, 7.9, 8.5 localization 2.8, 8.5 long division 6.4 long exact homology sequence S3 Maschke’s theorem 9.6 matrices 2.1, 4.4 maximal ideal 2.4, 8.3 maximal submodule 9.7 metric 7.9 minimal generating set 9.8 minimal left ideal 9.3 minimal polynomial 3.1 minimal prime ideal 8.4 modding out 5.7 modular law 4.1 module 4.1 modules over a principal ideal domain 4.6, 10.5 monic 10.1 monoid 1.1 monomorphism 1.3 morphism 10.1 Nakayama’s lemma 9.8 natural action 5.3, 6.3 natural map 1.3 natural projection 4.7, 7.5, 9.2, 9.4, 9.5, 10.5 natural transformation 10.3 naturality S3 Newton’s identities A6 nil ideal 9.7 nilpotent element 8.6, 9.7 nilpotent group 5.7 nilpotent ideal 9.7 nilradical 8.6 Noetherian modules 4.6, 7.5 Noetherian rings 4.6, 7.5 nonarchimedian absolute value 7.9 noncommuting indeterminates 9.8 nontrivial ideal 2.2 norm, 7.1, 7.3 normal closure, 3.5

page 11 of Endmatter

normal extension 3.5 normal series 5.6 normal Sylow p-subgroup 5.5 normalizer 5.2 Nullstellensatz 8.3, 8.4 number ﬁeld 7.1 objects 10.1 opposite category 10.1 opposite ring 4.4 orbit 5.2 orbit-counting theorem 5.3 orbit-stabilizer theorem 5.2 order 1.1 order ideal 4.2 orthogonal idempotents 9.6 Ostrowski’s theorem 7.9 p-adic absolute value 7.9 p-adic integers 7.9 p-adic numbers 7.9 p-adic valuation 7.9 p-group 5.4 perfect ﬁeld 3.4 permutation 1.2 permutation group 1.2 permutation module 9.5 polynomial rings 2.1, 2.5 polynomials over a ﬁeld 3.1 power sums, A6 preordered set 10.2 primary component A10 primary decomposition 8.6 primary ideal 8.6 prime element 2.6hbn prime ideal 2.4 primitive element, theorem of 3.5, 6.6 primitive polynomial 2.9 principal ideal domain 2.6 product 10.2 product of an ideal and a module 9.3 product of ideals 2.3, 8.5, 7.6 projection 4.7, 9.2, 9.4, 9.5, 10.5 projection functor 10.3 projective basis lemma 10.5 projective limit 10.9 projective modules 9.8, 10.5 projective resolution S4 proper ideal 2.2 Prufer group A10 pullback 10.6 purely inseparable 3.4 pushout 10.6,3 quadratic extensions 6.3, 7.2 quasicyclic group A10

page 12 of Endmatter

quasi-regular element 9.7 quaternion group 2.1 quaternions 2.1 quotient ﬁeld 2.8 quotient group 1.3 quotient ring 2.2 R-homomorphism on R 9.4 Rabinowitsch trick 8.4 radical extension 6.8 radical of an ideal 8.3 rank of a free module 4.4 rational integer 7.2 rational root test 2.9 rationals mod 1 1.1 (Problem 7), 10.4, 10.6, A10 reﬁnement 5.6 regular action 5.1, 5.2 regular n-gon 6.8 regular representation 9.5 relatively prime ideals 2.3 remainder theorem 2.5 representation 9.5 residue ﬁeld 9.8 resolvent cubic A6 restriction of scalars 8.7, 10.8 right adjoint 10.7 right cancellable 10.1 right derived functors S5 right exact functor 10.4 right ideal 2.2 right resolution S4 right-Noetherian ring 9.8 right-quasiregular element 9.7 right-semisimple ring 9.6 ring 2.1 ring of fractions 2.8, 8.5 Schreier reﬁnement theorem 5.6 Schur’s lemma 9.2 semidirect product 5.8 semigroup 1.1 semisimple module 9.1 semisimple ring 9.3 separable element 3.4 separable extension 3.4 separable polynomial 3.4 series for a module 7.5 simple group 5.1, 5.5 simple left ideal 9.3 simple module 7.5, 9.1, 9.2 simple ring 9.3, 9.5 simplicity of the alternating group 5.6 simultaneous basis theorem 4.6 skew ﬁeld 2.1 Smith normal form 4.5

page 13 of Endmatter

snake diagram S2 snake lemma S2 solvability by radicals 6.8 solvable group 5.7 spanning set 4.3, 6.9 special linear group 1.3 split exact sequence 4.7, 5.8 splitting ﬁeld 3.2 squaring the circle 6.8 standard representation of a p-adic integer 7.9 Steinitz exchange 6.9 Stickelberger’s theorem 7.4 subcategory 10.3 subgroup 1.1 submodule 4.1 subnormal series 5.6 subring 2.1 sum of ideals 2.2 sum of modules 4.3 Sylow p-subgroup 5.4 Sylow theorems 5.4 symmetric group 1.2, 6.6 symmetric polynomial 6.1 tensor functors 10.3 tensor product of matrices 8.7 tensor product of module homomorphisms 8.7 tensor product of modules 8.7 terminal object 10.1 Tor S5 torsion abelian group 8.7 (Problem 3), 10.2 torsion element 4.6 torsion module 4.6 torsion subgroup A10 torsion submodule 4.6 torsion-free module 4.6 trace 7.3 transcendence basis 6.9 transcendence degree 6.9 transcendental element 3.1 transcendental extension 6.9 transcendental number 3.3 transitive action 5.2 transitive subgroup of Sn 6.3 transitivity of algebraic extensions 3.3 transitivity of separable extensions 3.4 transitivity of trace and norm 7.3 transposition 1.2 trisecting the angle 6.8 trivial absolute value 7.9 trivial action 5.1, 5.2 twisted cubic 8.3 two-sided ideal 2.2 ultrametric inequality 7.9 underlying functor 10.3

page 14 of Endmatter

unimodular matrix 7.4 unique factorization domain 2.6 unique factorization of ideals 7.7 unit 2.1 universal mapping property (UMP) 8.7, 10.2 upper central series 5.7 valuation 7.9 valuation ideal 7.9 valuation ring 7.9 Vandermonde determinant A6, 7.4 vector space as an F [X]-module 4.1 Von Dyck’s theorem 5.8 weak Nullstellensatz 8.3 Wedderburn structure theorem 9.5 Wedderburn-Artin theorem 9.4 Zariski topology 8.1 Zassenhaus lemma 5.6 zero object 10.1

Basic Abstract Algebra: For Graduate Students and Advanced Undergraduates

Read more

Basic abstract algebra

Read more

Basic Abstract Algebra

Read more

Basic Abstract Algebra

Read more

Abstract Algebra (Graduate Texts in Mathematics)

Read more

Abstract Algebra (Graduate Texts in Mathematics)

Read more

Abstract Algebra (Graduate Texts in Mathematics)

Read more

A first graduate course in abstract algebra

Read more

Abstract Algebra

Read more

Abstract Algebra

Read more

Abstract Algebra

Read more

Abstract algebra

Read more

Abstract Algebra

Read more

Abstract algebra

Read more

Abstract Algebra

Read more

Algebra. Abstract and Concrete

Read more

Abstract algebra

Read more

Abstract algebra

Read more

Abstract Algebra

Read more

Algebra Abstract

Read more

Abstract Algebra

Read more

Abstract algebra

Read more

Abstract algebra

Read more

Abstract Algebra

Read more

Abstract algebra

Read more

Algebra. Abstract and concrete

Read more

Algebra. Abstract and Concrete

Read more

Algebra Abstract and Concrete

Read more

Algebra. Abstract and Concrete

Read more

Basic Homological Algebra (Graduate Texts in Mathematics)

Read more

Recommend Documents

Basic Abstract Algebra: For Graduate Students and Advanced Undergraduates

Basic abstract algebra

Basic Abstract Algebra

BASIC AB STRACT ALGEBRA P.B .Bhattacharya S .K.Jain S .R.Nagpaul (b i) - 1) j: () SECOND EDITION BASIC ABSTRACT...

Basic Abstract Algebra

Abstract Algebra (Graduate Texts in Mathematics)

Graduate Texts in Mathematics 242 Editorial Board S. Axler K.A. Ribet Graduate Texts in Mathematics 1 2 3 4 5 6 7 8...

Abstract Algebra (Graduate Texts in Mathematics)

Abstract Algebra (Graduate Texts in Mathematics)

Graduate Texts in Mathematics 242 Editorial Board S. Axler K.A. Ribet Graduate Texts in Mathematics 1 2 3 4 5 6 7 8...

A first graduate course in abstract algebra

Abstract Algebra

Graduate Texts in Mathematics 242 Editorial Board S. Axler K.A. Ribet Graduate Texts in Mathematics 1 2 3 4 5 6 7 8...

Abstract Algebra

Graduate Texts in Mathematics 242 Editorial Board S. Axler K.A. Ribet Graduate Texts in Mathematics 1 2 3 4 5 6 7 8...